Independence is easy to declare and expensive to organise. This week I saw what the organised version looks like, in Zurich, at the annual meeting of Group 39 and during a company visit to Neue Zürcher Zeiting - NZZ hosted by the Swiss news agency Keystone-SDA. The alliance takes its name from 1939, when Belga News Agency and like-minded agencies in Northern and Western Europe built their own network for exchanging facts while news provision elsewhere came under political control.

That institutional instinct has aged well. Editorial independence today depends on three pillars: ownership that cannot be captured, an economic model that does not depend on a single gateway, and technical infrastructure a company actually controls. NZZ offers a working example of the first two, and this week’s other signals show how contested the third has become.

The dispatch of this week is build around this visit to Zürich.

Large platforms are writing the terms on which trusted information is used, priced and distributed, through licensing programmes, crawler policy and assistant interfaces. Publishers, broadcasters and press agencies are answering with standards work, litigation, collective negotiation and internal capability building.

Courts treated the same question from three continents, with the Court of Justice clarifying copyright in social media posts, a German ruling on advertising liability, and new judicial guidance from China’s highest court. Regulators and lobbyists moved in Ottawa, Madrid and Tokyo. Inside the newsrooms, the practical work went on: transparency guidelines in the Netherlands, AI literacy for more than four thousand employees at Mediahuis, and verification tools built in Nigeria and Dubai.

Nine signals this week, in four chapters, running from ownership to protocol to daily editorial practice.

I. On independence as infrastructure

Two signals on what makes a news organisation difficult to capture.

29.1 — NZZ builds its digital future on ownership and reader revenue

NZZ reported growth in its core business and an improved operating result in a statement from the group’s corporate site. I’m mostly interested in the structure behind the numbers. According to the presentation of CEO Felix Graf during our visit, NZZ has roughly 2,700 shareholders, all of them individuals, with no single holder controlling more than one percent of the voting rights, and the title itself dates back to 1780, which places it highly ranked among the top 10 international quality titles editors still study.

Three operational facts explain the strategy. Around 75 percent of traffic arrives directly, which reduces exposure to declining search referrals, social distribution and substitution by AI assistants. Revenue concentrates on paying readers rather than volume, with digital subscriptions as the engine and selective diversification into adjacent media businesses and holdings as the hedge. Technology serves the editorial product through archive retrieval, style consistency and workflow efficiency, under human control, with none of the positioning of an AI laboratory.

This describes a conservative and sophisticated digital strategy: monetise scarce quality, spread the portfolio with discipline, keep technology instrumental. The value of this premium model lies in the demonstration that direct relationships with readers can be built and held, and that dispersed ownership makes editorial autonomy structural rather than aspirational.

Reflections How does NZZ’s resilience coming from its shareholder structure influences the markets willingness to pay for scarce quality? What would a mass-market publisher have to change to reach 75 percent direct traffic?

29.2 — Group 39 marks eighty years of shared independence

Group 39 now brings together eleven independent European news agencies: Belga, ANP, APA, Keystone-SDA, ANSA, dpa, PA Media, Ritzau, STT, TT and NTB. The alliance marked eighty years of cooperation and an expansion of its circle of members, a cooperation founded on the conviction that reliable factual exchange requires a network owned by its participants. The annual meeting in Zurich, hosted by Keystone-SDA, kept that agenda intact while adding the machine layer to it.

Ownership remains the underlying foundation of that model, and it is worth looking at the national picture. Hugues Rey’s survey of Belgian media from 1839 to 2026 describes almost two centuries of continuous consolidation with capital that has remained largely national. Concentration of that kind delivers scale and survival capacity; it also concentrates the decisions about which titles, languages and regions receive investment.

Press agencies occupy a specific position in this structure. They are critical information infrastructure for publishers, broadcasters, institutions and increasingly for machines, which gives their governance a public dimension that their balance sheets alone do not capture. An alliance founded in 1939 to protect factual exchange from political capture now faces a comparable task with commercial gatekeeping, and the instrument is the same: shared standards, shared costs, independent ownership.

Reflections Which shared capability would justify the next decade of an agency alliance, beyond the exchange of content? Where should the line fall between cooperation on infrastructure and competition on products?

II. On who sets the rules and the price

Two signals on licensing programmes, crawler policy and the emerging market structure.

29.3 — Google puts a value on publisher content and keeps the scoring

Google has begun rolling out an AI licensing programme for publishers built on the principle of paying for value, with compensation tied to content that contributes significantly to its systems. Two questions decide what the programme is worth to a publisher: what counts as significant contribution, and who calculates it. Participants at the Digiday Publishing Summit discussed the same dependency under the heading of Google Zero, while OpenAI signed a Canadian agreement with Villager Media covering a limited set of titles.

Individual agreements of that type reward the publishers who arrive early and leave the rest of the market waiting. No technology company can conclude a bilateral partnership with every news organisation in every language, which makes the bilateral approach an opportunistic market entry with a hard ceiling. At ISOJ, the discussion turned to the consequence: publishers now face a new intermediary layer as AI reshapes the business, an interface that sits between the newsroom and its audience and sets the terms of access.

Media executives have started to say this in public, with senior leaders sounding the alarm about AI’s effect on their businesses. The economics are conventional: a scarce input acquires a price when the buyer needs it and the seller can withhold it. Pricing power therefore depends on collective repertoire, auditable measurement and the credible option of saying no, and every licensing programme designed by a buyer should be read as a proposal about who keeps the ledger.

Reflections What auditable evidence would a publisher need before accepting a value-based payment formula designed by the buyer? How does a market reach a workable price when the largest buyers negotiate one publisher at a time?

29.4 — Cloudflare and SPUR turn access into a technical question

Cloudflare published new work on accountability for mixed-use AI crawlers and announced a mechanism intended to end the tradeoff between being indexed for search and being used for AI training. In the past, publishers were tempted to accept training as the price of discoverability. Separating the two purposes at the level of the request gives rights holders a choice they could previously express only by accepting everything or blocking everything.

The model of Cloudfare shows one strategic problem. Websites are built for human readers; machines are better served through an API, where a publisher controls identification, volume, purpose and price by design. Crawling remains a permission problem solved after the fact, while an interface is a commercial arrangement agreed in advance, and the second is easier to audit and to invoice. That is where standards work becomes decisive, and SPUR’s technical lead Alex Springer has described the coalition’s effort to build an economic model for publishers through machine-readable rights, identification, metadata and telemetry.

A shared technical layer of that kind has one condition attached: it must stay separated from pricing and commercial terms, so that competitors can cooperate on protocol while competing on value. Infrastructure built jointly and priced individually is how other regulated markets solved the same coordination problem. Rights that machines can read are worth more than rights that require a lawyer to interpret, because only the first can be enforced at the speed at which content is consumed.

Reflections If machine access moves to APIs by default, why would a publisher still allow crawling? Which functions belong in a shared industry standard, and which must remain each company’s own commercial decision?

III. On the law finding its equilibrium

Two signals on courts, filings and legislative preparation across several jurisdictions.

29.5 — Court filings and executive statements reshape the copyright debate

EANA reported that unsealed court filings indicate the collection of news content for AI training was deliberate and went uncompensated, a finding drawn from documents produced in litigation rather than from press statements. Reuters separately reported that quotes from OpenAI and Microsoft executives may weaken their defence in copyright cases brought by news organisations. Evidence of intent carries weight in any assessment of good faith, and internal communications have become part of the factual record.

In Europe the framework is droit d’auteur and the CDSM Directive, which means the relevant question concerns authorisation and remuneration for specific acts of reproduction and extraction. The Court of Justice added a further element by clarifying the conditions for copyright protection of social media posts, a ruling that matters for every organisation using social content in reporting and for every system trained on it. Originality thresholds decide what may be reused at scale.

Reflections How should a negotiator value machine rights and based on what permitted use? What evidentiary standard should apply and why should publishers accept intent, rather than volume, as the attribution key?

29.6 — Three jurisdictions set different limits on AI and platform liability

China’s highest court issued new judicial opinions on AI-related disputes, drawing explicit red lines, giving lower courts a framework where legislation remains incomplete. In Germany, a court held Meta liable for fraudulent advertisements on Instagram and Facebook, extending platform responsibility for commercial content carried on its systems. Japan, meanwhile, secured budget funding to examine a system modelled on South Korea’s AI-supported response to online copyright infringement.

Canada shows the same movement at the level of policy preparation, where media organisations raised AI and the protection of journalism in meetings with Ottawa. Four jurisdictions, four instruments: judicial guidance, civil liability, enforcement technology and legislative lobbying. Each one changes the compliance surface for companies operating internationally, and each one creates a precedent that counterparts will cite in the next negotiation.

Reflections Which of these instruments will other jurisdictions copy first, and why that one? How should a European rights holder enforce a licence when conditions differ in other markets?

IV. On the newsroom as the product

Three signals on strategy, trust and internal capability.

29.7 — Publishers redefine what a newsroom sells

Axel Springer’s chief executive Mathias Döpfner told dpa that he intends to develop the company into a creator platform, in an interview published by HORIZONT. The Reuters Institute described the same movement across the industry as a major pivot by newspaper groups towards news creators, and the underlying logic is straightforward: individual voices travel through algorithmic distribution more easily than institutional brands, and attention has become the scarce input.

Two other decisions this week show the cost side of that strategy. Reach is cutting 220 editorial jobs, reported by The Next Web and by The Guardian in the context of AI summaries and search referrals, while News Corp has started generating local news stories with AI assistance. Reuters took the opposite route to differentiation by extending its Pure News, Straight from the Source brand campaign with two new videos, and the Reuters Institute gathered five takeaways on how publishers are rethinking audience from its Future of Media Technology conference.

Institutions deliver something a creator cannot: verification capacity, legal protection, correction procedures and continuity when a story becomes dangerous to publish. The strength of a quality newsroom lies in that collective apparatus, and it survives the departure of any individual byline. A company that markets personalities while reducing editorial capacity sells the signal and reduces the newsroom that produces it, which looks more like a budget decision disguised as a distribution strategy.

Reflections What does a publisher own in a creator model when the audience follows the person? Which editorial functions must remain institutional for a brand to keep charging a premium?

29.8 — Verification becomes an operational discipline in more markets

A report documented how actors in the Democratic Republic of Congo used Claude to produce fabricated news articles, and The New York Times examined how AI chatbots are shaping election misinformation. Institutions responded with instruments of their own. South Africa’s electoral commission launched a code of conduct to limit disinformation ahead of elections, Dubai released Saraab, an open source model for detecting deepfake videos, and Nigeria’s Centre for Journalism Innovation and Development built Dubawa AI to help journalists and the public verify claims across Africa.

The security dimension belongs to the same picture. The United Kingdom, the United States and the Netherlands issued a joint advisory on Iranian spyware, a reminder that journalists and their sources are surveillance targets and that newsroom security is part of editorial capacity. Meanwhile 404 Media argued that AI agents are already reshaping the internet, which changes the arithmetic of verification at the level of infrastructure.

Capability is spreading outward from the traditional centres of media technology, and Lagos, Dubai and Johannesburg now produce tools and rules that news organisations can use. Verification has acquired the character of a permanent operating function with staffing, tooling and budget, comparable to information security a decade ago. Trust is produced at a cost, by people and systems that have to be paid for, and it stays scarce for exactly that reason.

Reflections Which verification capacity should a news organisation build internally, and which should it source from shared infrastructure? How do editorial teams document verification in a way an audience can actually check?

29.9 — Mediahuis measures AI adoption by the drop in confidence

Mediahuis made AI literacy and adoption mandatory for more than 4,000 employees across Belgium, the Netherlands, Luxembourg, Germany and Ireland, and the group’s head of AI strategy Ana Jakimovska has published what those 4,000 people taught the company about AI adoption. The programme ran in three phases with two full-time staff and five part-time colleagues, reached 78 percent participation, and included the Board in the same training. Self-reported confidence stood at 3.85 out of 5 after the introductory phase, then fell to around 3 once employees applied the tools to their own workflows, and the third phase has produced 177 use cases now moving through prioritisation, build and measurement. It fits a wider repositioning that Jakimovska earlier described as a move from experimentation to adoption.

That decline in confidence carries the strategic lesson. Employees moved from asking whether AI can perform a task to asking which part of the process AI should handle, which part requires human judgement, and how the two connect; Jakimovska concludes that rising confidence on its own measures very little, and that professional hesitation about quality, accountability and expertise often represents judgement at work. For newsroom leaders this reframes the investment case. Licences and pilots are the cheap part of the programme, and the expensive part is redesigning workflows around verification, with the cost falling on the same editorial capacity that produces the value.

The Mediahuis evidence was published in the same week that Dutch media organisations agreed joint guidelines for AI transparency and that comparative research on citizens’ attitudes toward AI in journalism and authenticity across six countries found broad agreement on the human element as the source of authenticity, with more optimism in South Africa and Brazil and more scepticism in Western markets. These three pieces of one governance question can be seen togheter: what the newsroom does, what it discloses, and what the audience accepts. Internal literacy makes disclosure credible, because an organisation can only account for a process it understands, and the number that matters is the one an editor can defend.

Reflections Which measure of AI adoption would survive scrutiny from an editor, a works council and an auditor at the same time? When trained staff conclude that a use case adds little value, how does a management team make room for that verdict?

Conclusion

The weeks dispatch is about who declares independence and who organises it. Declarations are inexpensive and appear in every mission statement. Organisation shows up in a shareholder register, in the share of traffic that arrives directly, in a standard that machines can read, in a training budget for four thousand people, and in the willingness to litigate while negotiating.

NZZ and Group 39 supply the constructive counter-example to the broader anxiety in the sector. Dispersed ownership, revenue from readers who choose to pay, cooperation on infrastructure with competition on products, and technology used as an editorial instrument. None of that requires a company to outspend the largest technology firms; it requires a company to control the parts of its own value chain that determine price.

The pricing question returns in every chapter. A licensing programme designed by the buyer, a crawler policy set by an intermediary, an assistant interface positioned between a newsroom and its audience: each shifts the power to define what a contribution is worth. Courts are establishing the boundaries of permitted use, and negotiations inside those boundaries will establish the price.

Trusted information has the economics of any scarce, costly product. It is produced by trained people working inside institutions that verify, correct and stand behind the result, and it stops being produced when nobody pays for the apparatus. The instructive detail this week came from Mediahuis, where confidence in AI fell as understanding rose. That is what competence looks like in a new domain, and any board that reads falling confidence as failure will fund the wrong programme.

An independent news organisation resembles a building with three load-bearing walls: ownership, revenue and protocol. Remove one and the structure still stands for a while, which is precisely what makes the decision easy to postpone.

Patrick Lacroix writes Cyber Territories in a personal capacity. AI tools are used for research and drafting; the author retains full editorial responsibility.

Ireland chairs the Council of the European Union until the end of December, with competitiveness, values and security as the three priorities of its presidency. For news organisations, those priorities invite a practical question: who defines the conditions under which AI companies can use journalism, and who checks that they respect them?

The Irish agenda combines AI adoption, copyright and the competitiveness of European businesses, while its programme also prioritises child safety and privacy-preserving age verification. Ireland intends to advance a review of the CDSM directive and seek agreement on the broader Digital Omnibus by the end of 2026. These objectives require decisions about access, evidence, responsibility and cost. Each decision affects the organisations that produce information and the people who depend on it.

On 14 October, NewsBrands Ireland and the Irish Recorded Music Association will bring copyright and human creativity into that discussion at a conference in Dublin devoted to licensing, trust and effective enforcement. News organisations have an opportunity to explain what such a market requires, from permission and attribution to reliable usage records and a sustainable price. The discussion will be more useful if the sector can describe how those requirements work in practice.

This edition considers those requirements alongside newsroom interoperability, AI discovery, media intelligence and synthetic influence. My working proposition is that standards deserve the same executive attention as contracts: they determine what a buyer can request, what a supplier can prove and how the parties resolve a dispute. The sector’s negotiating position depends partly on its ability to specify its own terms.

European Commission President Ursula von der Leyen and Taoiseach Micheál Martin at University College Cork during the College of Commissioners’ visit at the start of Ireland’s Council presidency in July 2026.

Photo: PA via Belgaimage.

I. On who writes the rules for training and safety

Governments, rights holders and developers approach the same negotiations with different legal obligations and commercial interests.

28.1: Ireland puts licensing on the agenda as Washington argues for broader training rights

The Dublin copyright conference on 14 October will examine licensing and enforcement under the Irish Council presidency, with NewsBrands Ireland and IRMA as joint hosts and the Department of Enterprise, Tourism and Employment endorsing the event. Its announced speakers include Henna Virkkunen, Peter Burke, Niamh Smyth and Eleonora Rosati, bringing European policy, Irish government and copyright expertise into the discussion. The conference can help define the sector’s priorities; its participants will still need legislators, regulators, courts and negotiating parties to act on them.

In Washington’s account of the problem, AI developers need broader access to training material: Reuters reports that Commerce Secretary Howard Lutnick urged G20 members to embrace the American copyright exception, while the Justice Department supported OpenAI’s position in the New York Times litigation. Those interventions express the US administration’s position and leave the courts to decide the disputes. The G20’s final ministerial statement recognises copyright protection, prior and express consent, and applicable exceptions, while leaving AI-related copyright questions to each member’s established legal processes.

Europe has its own legal basis: Article 53 of the AI Act requires general-purpose model providers to maintain a copyright compliance policy, including respect for rights reservations under the CDSM directive. The sector also has a practical point of reference in SPUR’s work on publisher control, licensing access and standards for usage reporting. Courts establish the legal boundaries, and negotiations establish workable prices; common technical standards should help both parties demonstrate compliance without dictating each publisher’s commercial terms.

Reflections Which licensing requirements should European news organisations bring to Dublin as a common proposal? How can shared technical standards strengthen smaller rights holders while preserving independent pricing and competition?

28.2: Amodei calls for stronger safeguards while Europe already requires systemic-risk controls

The Next Web reports that Anthropic’s Dario Amodei wants developers to allow more time for alignment, safeguards and external evaluation as model capabilities increase. His proposals include outside evaluators with extensive access to company systems, an approach that raises practical questions about independence, confidentiality and authority. In Britain, meanwhile, the government has rejected a proposed national AI “kill switch”, arguing that domestic restrictions would offer limited protection against development and misuse elsewhere.

For the EU, several relevant duties already have legal force: Article 55 requires providers of general-purpose AI models with systemic risk to evaluate models, document adversarial testing, assess and mitigate systemic risks, report serious incidents and protect their models and infrastructure against cyber threats. The general-purpose model rules have applied since 2 August 2025, although Article 111 gives providers of models placed on the market before that date until 2 August 2027 to comply. The AI Omnibus that entered into force on 27 July 2026 postponed high-risk system deadlines, while the Commission’s current implementation guidance retains the general-purpose model timetable. Amodei’s proposed evaluator arrangements require their own assessment; Article 55 establishes substantive duties without prescribing every feature of his proposal.

The practical test concerns the evidence that providers submit: Euractiv reports that the Commission demanded precise information about corrective measures after OpenAI notified it of the DseWiki incident. News organisations procuring agentic systems should ask for documented testing, incident notification commitments and explicit authority to suspend a workflow. A provider’s assurance has operational value when the customer can identify what the provider tested, what failed and who must respond.

Reflections What evidence should a newsroom require before granting an AI agent access to editorial systems or confidential material? How should regulators and customers evaluate a provider’s corrective measures while protecting legitimate security information?

II. On documenting origin and preserving editorial context

News organisations need standards that describe both the material they receive and the decisions they make with it.

28.3: IPTC advances provenance practice as Press Gazette introduces human-authorship checks

The IPTC Photo Metadata Conference examined practical implementations of C2PA, identity assertions and metadata across news and imaging workflows. IPTC also introduced Metawatch, a public service monitoring metadata use across the news industry, including content credentials and AI opt-outs. That work gives the sector a way to examine whether publishing tools preserve the information that standards promise.

Press Gazette has taken a related commercial initiative with Made by Humans, which combines Pangram and Identifai checks in a platform developed with Flip-Pay. Press Gazette says the service checks text, images, video and audio before publication and records reviews for audit purposes. Editors considering such a service should examine its error rates, appeal procedures and definition of human authorship, particularly when a journalist uses AI for translation or editing.

The C2PA specification distinguishes verifiable provenance from a judgement about the truth of the content. A newsroom should make an equally precise distinction between evidence of origin, an assessment of AI involvement and editorial verification. Each answers a different question, and a trustworthy label should tell the reader which question the publisher actually checked.

Reflections What should a publisher promise when it displays a human-authorship or provenance label? How can newsrooms preserve useful evidence of origin while protecting confidential sources and allowing legitimate editing?

28.4: AP and Reuters pursue shared context as suppliers automate more newsroom tasks

Journalism.co.uk reports that AP and Reuters want a common language through which AI tools can exchange context from newsgathering to publication. Buyers should follow the initiative’s development and ask for a published specification and evidence of adoption. The essential question concerns which editorial information every participating tool must preserve.

In its IBC announcement, nxtedition describes a demonstration that uses Reuters material to develop a story, select video, prepare graphics and assemble a programme, with journalists and directors reviewing the output. Separately, Reuters has replaced Getty as ITN’s archive agency partner, with plans to digitise another 10,000 hours of footage and improve discovery through Reuters Imagen metadata and Reuters Connect. These announcements show suppliers combining content, description, licensing and production tools into broader services.

Press agencies have a particular responsibility here as information infrastructure serving many other news organisations. A useful shared specification should preserve attribution, rights restrictions, corrections and editorial status whenever software transforms an item. Publishers and public broadcasters should help define those requirements before procurement decisions make a supplier’s internal conventions their own operating standard.

Reflections Which editorial and rights information must every tool preserve when it transforms agency material? How can customers obtain interoperable workflows without depending on one supplier for content, metadata and production?

III. On discovery, measurement and negotiated access

Audience strategy and commercial measurement increasingly depend on how intermediaries select and represent original reporting.

28.5: Publishers reorganise audience teams around AI discovery

Digiday reports that USA Today plans an audience and digital production team of 23 to 30 people as it responds to changes in search and distribution. At the New York Times, Christine Liang has taken responsibility for AI and off-platform discovery, following the creation of a comparable leadership role at the Washington Post. The organisational choices matter because they assign people and budgets to a changing relationship with audiences.

A publisher needs to know which forms of visibility produce a reader relationship and which mainly improve an intermediary’s answer. Audience teams should therefore compare AI citations and referrals with subscriptions, repeat visits, newsletter engagement and other evidence of sustained value. Optimising for an appearance in an answer carries a different commercial logic from persuading a reader to return.

Public policy also influences discovery: Nieman Lab describes Australia’s proposed Digital Duty of Care, including greater user control over recommendations and personalisation. For European news organisations, that debate suggests a useful test for distribution policy: how much choice can the audience exercise, and what access can a publisher retain? A durable audience strategy should accommodate changing interfaces while continuing to give people reasons to seek out the newsroom.

Reflections Which measures would demonstrate that AI discovery creates lasting value for a publisher? How should audience teams divide investment between intermediary visibility and direct reader relationships?

28.6: Media-intelligence suppliers compete on AI visibility, local context and contract terms

Onclusive’s Unified Platform announcement combines monitoring, measurement and communications workflows, including GEO Analytics for visibility across AI search platforms. The company also describes a shared content foundation that includes directly licensed and paywalled material. Its offer gives communications teams another object to measure: how AI services represent an organisation when users ask questions.

Truescope plans to enter Australia on 1 October with a maximum twelve-month contract term, no automatic renewal and cited AI outputs. In Nigeria, Punch reports that FlashNarrative claims capabilities for local language, code-switching and cultural context alongside its brand-intelligence metrics. These announcements describe the suppliers’ offers; customers still need evidence of coverage, accuracy and commercial performance.

The three approaches suggest a broader procurement agenda: the object of measurement, the quality of interpretation and the customer’s freedom to change supplier. Buyers of AI-visibility analysis should require a reproducible method, including the models, prompts, languages, dates and sampling assumptions behind each result. A metric earns value when the customer can understand its limits and use it to support a defensible decision.

Reflections What minimum methodology should buyers demand before treating AI visibility as a management metric? How should media-intelligence contracts allocate responsibility for source rights, analytical errors and changing model behaviour?

28.7: Indian Express negotiates attribution and editorial control with OpenAI

The Indian Express Group and OpenAI have announced a partnership covering current and archived journalism across seven languages. The publisher says selected summaries and excerpts in ChatGPT will identify the original source and include direct links, while the group retains independent editorial control. The agreement offers a concrete example of a publisher specifying how an AI service should represent its journalism.

The public announcement describes attribution, editorial safeguards and possible uses in archive discovery, research and translation, while leaving payment and detailed usage terms undisclosed. We can assess the announced design and reserve judgement on its economics. For any publisher, the commercial test should include the cost of producing the journalism, the scope of permission and the evidence needed to reconcile actual use with the contract.

European news organisations can use such announcements to identify questions for their own negotiations. Training, retrieval, display, translation and internal newsroom assistance deserve explicit treatment because each permits a different activity. A licensing market will work better when the parties can describe those activities consistently and negotiate the price of each permission on informed terms.

Reflections Which permissions should a publisher price separately when negotiating with an AI provider? What reporting and audit rights would allow both parties to evaluate the partnership beyond attribution and referral counts?

IV. On platforms financing newsroom capacity

Training and technical support can create valuable capabilities, alongside dependencies that managers should assess at the outset.

28.8: OpenAI and Google support journalism training and product development

OpenAI has announced more than 400 ChatGPT Edu subscriptions for students and faculty through CUNY’s Newmark J-School and Medill’s Knight Lab. Its broader programme also includes access and API credits through the American Journalism Project, support for AI engineering fellows through Lenfest, and its WAN-IFRA partnership. The initiatives combine access to tools with the people and training needed to use them.

In Ukraine, WAN-IFRA, AIRPPU and OpenAI plan a Newsroom AI Catalyst programme for ten organisations from 17 September, following masterclasses that began in August. In Latin America, O Globo and nine other newsrooms are participating in an eight-month News Product Alliance programme for Grupo de Diarios América, with Google News Initiative support, to develop AI products for younger audiences. Both programmes address practical capacity, although their announcements offer limited grounds for judging lasting results.

Newsrooms should welcome useful support while defining the capabilities they intend to retain when the programme ends. Managers should establish who owns the resulting tools, which data a partner can access, how staff can transfer their skills and what the service will cost after credits expire. A successful partnership should leave the newsroom better able to make its next technology decision on its own terms.

Reflections Which capabilities should a newsroom retain independently after platform support ends? How should journalism schools and industry associations evaluate the influence of technology sponsors on their curricula and product choices?

V. On synthetic influence and the audience’s ability to judge

Effective safeguards must address commercial incentives, enforcement and the public’s capacity to assess information together.

28.9: Investigations expose synthetic influence while young audiences turn to AI for news

ABC reports Reset Tech’s findings of a network impersonating Australian politicians, with links to an influencer business in Sri Lanka. The researchers found monetisation on roughly one in ten of the network’s Facebook pages, AI labels on only one in three examined posts and deletion in one in 33 cases. Meta told ABC that it was reviewing the identified pages and would act against policy violations. The immediate governance questions concern the incentives for producing such material and the speed and consistency of enforcement.

Semafor reports that YouTube terminated twenty channels after an investigation into “ghost creators”, human presenters reading AI-generated scripts while appearing to offer independent commentary. In Mali, RFI reports the Timbuktu Institute’s findings on synthetic characters and propaganda, alongside its warning about the potential for multilingual recruitment. THISDAY’s examination of Nigeria’s approaching elections adds an important distinction: manipulated political material includes both AI-generated content and older editing techniques. Verification should examine the claim and its context as carefully as the production method.

The audience research gives these cases particular relevance: Western Sydney University reports that more than a third of surveyed Australians aged ten to seventeen use generative AI for news, while fewer than half feel confident distinguishing fact from fiction online. The same study finds that three-quarters believe news organisations have little understanding of their lives. For publishers and public broadcasters, verification and relevance therefore belong in the same strategy. Young people need dependable information that recognises their concerns and gives them reasons to examine its evidence.

Reflections How should platforms demonstrate that their monetisation and enforcement systems reduce the incentive to produce deceptive political content? What would news organisations change if they treated young people’s sense of exclusion as seriously as their exposure to misinformation?

Conclusion

The Irish presidency gives the sector a practical occasion to connect copyright, AI governance, competitiveness and public trust. A useful contribution from news organisations would specify what they need to produce valuable journalism and make it available responsibly. That contribution should include a commercial model, technical requirements and evidence of how the arrangements serve audiences.

A standard resembles the schedule attached to a contract. It defines the fields, permissions and records that determine whether the parties can perform what they promised. If news organisations leave that schedule entirely to others, they accept another party’s assumptions about what matters and what deserves a price.

The Dublin conference on 14 October explicitly links a licensing market with enforceable copyright protection. Publishers, public broadcasters and press agencies should bring proposals that reflect their different roles, including agencies’ responsibility for the verified material that many organisations reuse. Shared definitions of permission, provenance, corrections and usage can support that work while preserving independent editorial judgement and commercial negotiation.

The public ultimately needs to know who produced the information, why it merits attention and who accepts responsibility for correcting it. Europe’s news sector has experience worth contributing to each of those questions. The task for this presidency is to make that experience count in the rules and standards that institutions and companies actually apply.

This week therefore we will draw a picture of what the members of that coalition and their media clients are actually doing. The Danish agency Ritzau put official emergency messages directly onto citizens’ phones. The Taiwanese press agency begins selling machine access to thirty years of archive by monthly subscription. The American agency AP moves video production to India. A Swiss publisher signed a group-wide enterprise agreement with OpenAI while avoiding to license its journalism. A Regional Court in Munich decided that an AI model had memorised protected works, and two American newspapers filed a fresh copyright claim in New York.

These signals are the signs of an economy under construction. Content is becoming a licensed, measured and machine-readable, distribution is becoming a question of who controls the access layer, and the value of a newsroom is defined by what it can prove about provenance, rights and accuracy.

Markus Knall of Ippen.Media brought back a useful term from an AI and media exchange hosted by Mediahuis in Brussels, the barbell model: AI splits the media market into boutique journalism at one end and large efficient services at the other, while the thin middle integrated both extremes as a lean operational core. This dispatch uses that hypothesis as a reading frame, and treats the sector news from EANA and SPUR members as a barometer of where the weight is settling.

I. On the sector organising itself

One decision that changes the level at which the industry negotiates.

27.1 — EANA takes the agency case into a standards coalition

EANA, founded in Strasbourg in 1956 and headquartered in Bern, has become an affiliate member of the SPUR Coalition, whose members include the BBC, the Financial Times, Sky News, Guardian Media Group, Telegraph Media Group, Mediahuis and the Associated Press. The News Media Coalition confirms that the arrangement gives EANA’s 33 agencies reduced-rate SPUR membership and a seat in standard-setting on content licensing and usage transparency, and the move was also picked up by the Slovak agency TASR. Secretary General Alexandru Giboi framed the affiliation as bringing the perspective of Europe’s leading agencies into what he called one of the most consequential debates of today.

The strategic content of this decision lies in the word standards. Licensing negotiations between a single agency and an AI model do little more than setting a price. A shared reporting standard for how generative applications use news content produces something more durable: a measurable unit of consumption, an audit trail, and a basis on which many rights holders can price the same behaviour consistently. For agencies, which supply media, companies, institutions and intermediaries rather than consumers, that measurement layer is the commercial foundation of the entire business.

There is a second effect worth naming. Agencies and publishers have overlapping interests and genuinely different positions, since an agency sells rights into the market that publishers themselves resell downstream in altered form. Putting both inside one standards body creates options for those positions to be reconciled in technical definitions instead of in litigations and opportunism. That is how infrastructure industries have historically settled: the specification comes first, and the tariff follows.

Reflections

Which decisions should a standards coalition be allowed to take on behalf of its members, and which must remain with each members own board?

How can we know, twelve months from now, whether shared measurement has improved the price of news content or merely standardised it?

II. On press agencies as critical information infrastructure

Three signals in which agencies define the layer through which official, verified and machine-readable information reaches the market.

27.2 — Ritzau places official messages directly in citizens’ hands

On 2 September, Ritzau launched DK Update, an app carrying operational messages from all Danish police districts, the Prosecution Service and Emergency Services Copenhagen, available in the App Store and Google Play. The service extends an existing Ritzau product that already distributes authority messages to virtually every Danish newsroom, and the agency states the ambition of including as many authorities as possible so that citizens can consult one place. The product manager describes the launch as a response to demand from media and journalists for direct notification when new information becomes available.

This is a business-to-consumer product built on a business-to-business asset, and that combination is the disrupting part. The value in DK Update is the pre-existing authenticated pipe between public authorities and the agency, with its verification discipline, its editorial standards and its uptime obligations. Agencies such as Belga operate that pipe exclusively towards media, companies, institutions and intermediaries; Ritzau decides to use the same infrastructure to serve the public directly, concluding that reliability, and not reach, is the scarce good.

The broader pattern concerns the definition of a press agency in a period of synthetic content. An institution that can guarantee the origin of an official message performs a function closer to a utility than to a publisher. Governments and regulators tend to notice such functions late, and to regulate them only after a failure.

Reflections

When an agency distributes official messages straight to citizens, what editorial responsibility does it take on beyond accurate transmission?

Where should the line run between a public authority’s own communication channel and an independent agency’s infrastructure?

27.3 — Agencies start selling access to machines, by subscription and by integration

Taiwan’s Central News Agency launched a subscription-based Model Context Protocol service on 31 August, letting applications such as ChatGPT, Claude and Grok query nearly five million news stories from more than thirty years, 3.5 million photographs spanning a century and public data from around 150 government agencies, with source citation built in, at NT$200 a month. AFP took a different route, announcing with Dalet an integrated newsroom experience in which AFP video arrives inside Dalet Flex with its scene-by-scene editorial metadata, shot composition, locations, soundbites and linked articles preserved. Belga News Agency, where the author of this newsletter serves as CEO, announced a partnership with FirstHour.ai covering the crisis communication chain from incident detection and AI-assisted drafting through approvals to distribution via Belga.share and post-crisis analytics, with NIS2 and DORA documentation as the reference frame.

Three products, one commercial logic: the customer is a system , and the deliverable is structured, rights-cleared, attributable content with a meter attached. CNA prices machine access explicitly and states a public-interest reason for it, namely reducing inaccurate information about Taiwan by supplying verified local reporting and Traditional Chinese terminology. AFP prices the preservation of editorial context, which is what makes archive material usable at speed. Each approach converts metadata discipline, long treated as a cost centre, into the reason a customer pays.

Ana Jakimovska of Mediahuis captured the shift in Knall’s account of the Brussels session mentioned earlier: B2C becomes B2A2C, and B2B becomes thus B2A2B, with content read first by machines and only afterwards by people. Agencies have an advantage here, because they have priced structured delivery to third parties for a century. The scarce asset is a catalogue whose provenance can be demonstrated line by line, and there is no such thing as a free lunch in producing one.

Reflections

What is the right unit of pricing when the customer is an agent: the query, the token, the citation, or the commercial outcome it supports?

How much of a newsroom’s metadata discipline should be sold as a product, and how much retained as a competitive advantage?

27.4 — Agencies rebuild production and commercial bases

The Associated Press is laying off 20 United States video production staff and moving the work to a centralised hub in Noida, with a further 10 American photo editors and 10 London video staff identified, following the elimination of roughly 8% of its global workforce in 2024; executive editor Julie Pace says the hub will scale video and photo capabilities while improving speed and quality, and the News Media Guild questions how well staff outside the country will know American topics and regions. NTB has appointed Lars Bratsberg to lead commercial growth, a hire from Spond with nine years at Google behind him, as the agency sells news, communication services and technology products into 22 countries and reorganises its picture agency of more than 500 million assets. Alongside these, dpa is continuing its Super Searchers information-literacy programme with BAGSO after training more than a thousand mostly library staff, Kazinform and the Albanian Telegraphic Agency signed a content and training memorandum on 1 September, and Ukrinform is publishing a book drawn from a year of commander interviews for the Lviv Publishers’ Forum.

Visual output represents more than 80% of AP’s production, which makes the cost per finished asset the decisive variable in the agency model; centralising that work reduces unit cost and transfers local knowledge risk to the editing layer. NTB’s hire points the other way, towards revenue: an agency that sells software into 22 countries needs commercial leadership from technology markets rather than from print sales. Both are the same strategy at different ends of the profit and loss account.

The pattern to watch is specialisation. dpa monetises trust through media literacy, Ukrinform monetises a documentary archive through publishing, Kazinform buys reach through bilateral agreements. Agencies that identify which scarce capability they own tend to leverage it; those that continue to create undifferentiated volume compete against the marginal cost of generation, which is heading towards zero.

Reflections

How far can visual production be centralised before the local knowledge that made it credible stops arriving with the footage?

Which capability inside your own organisation is genuinely scarce, and what evidence would convince a client to pay a premium for it?

III. On governance at the base of the system

Two agencies and one publishers’ association show how quickly institutional independence becomes an operational question.

27.5 — Independence is tested at the level of contracts and appointments

The interim leadership of Hungary’s public media fund MTVA announced it will file a criminal complaint after an internal review concluded that the organisation had been directed outside its formal governance, with a professional consultant, contracted as an adviser since 2019 allegedly deciding personnel and operational matters; staff reportedly referred to an invisible fifth floor in a four-storey building. In Bulgaria, BTA director general Kiril Valchev is taking unpaid leave to run as a vice-presidential candidate, saying he had not sought the nomination and citing more than twenty years of legal practice. In Hungary, the publishers’ association MLE announced two new members while acting as a co-regulatory participant in the committee assessing candidates for the presidency of the Media Council and nominating a member to the Independent Public Media Board. Ukrinform, meanwhile, hosts a round table on 11 September on the ethical use of AI in Ukrainian newsrooms, building on a policy template already adopted by seven outlets.

Editorial independence is usually discussed as a principle and is always at peril to be lost as a procedure. The MTVA allegations describe informal authority operating through advisory contracts and message traffic, which is precisely the space that governance codes tend to leave undefined. The BTA case raises the mirror-image question of what an agency chief executive’s public candidacy does to the perceived neutrality of the wire, leading to unpaid leave as an answer.

The Ukrainian round table shows the constructive version of the same challenge. Written newsroom AI policies, adopted outlet by outlet and tested against legal and ethical requirements, convert a contested principle into an auditable procedure. Institutions that document who decided what, and when, are considerably harder to capture.

Reflections

What safeguards would make advisory contracts inside a public media organisation transparent enough to be verifiable from outside?

Which political roles should an agency’s chief executive be able to seek with unpaid leave as available remedy?

IV. On publishers testing the barbell

Signals from SPUR members and the wider publishing market, read against the Barbell hypothesis.

27.6 — Ringier separates AI tooling from content licensing

Ringier and OpenAI have launched a multi-year collaboration providing long-term access to OpenAI’s models and a phased rollout of ChatGPT Enterprise across the group under European data storage, covering employee enablement, software development, marketplaces, editorial work and corporate functions. The agreement explicitly excludes any licence of Ringier’s content, and chief executive Marc Walder states that technology supports people and does not replace journalistic responsibility. SRF’s analysis examines the tension with Walder’s earlier description of unauthorised use of journalistic content by AI as the greatest theft in the history of journalism, and points out that Switzerland has no binding rules on declaring AI use or its extent in editorial products.

The construction here is more precise than the commentary suggests. Ringier is buying enterprise capability while withholding the asset it intends to price separately, which is a rational sequencing decision for any rights holder that expects a licensing market to mature. The exposure lies in the operational detail: enterprise deployments involve prompts, documents and workflows that carry editorial knowledge, and the contractual boundary around that material determines whether the separation holds in practice.

For the sector, this is the template most large European groups will examine. Buy tooling under European data terms, keep the content licence as a distinct negotiation with its own price and its own machine-readable reservation of rights, and publish the internal rules on declaration before a regulator writes them.

Reflections

Which contractual guarantees would satisfy you that enterprise AI deployment leaves your content licensing position intact?

What should readers be told about AI use in an editorial product, and who is best placed to set that standard?

27.7 — The middle of the market is where the pressure lands

Sipa Ouest-France has identified 12 million euro of payroll savings to avoid a redundancy plan at France’s most widely circulated daily, with the sale of its Paris headquarters and radio stations expected to raise more than 14 million euro by the end of 2026. Bonnier News is consolidating 35 salespeople into a single hub for print advertising across Dagens Nyheter, Expressen and Mitt i. Sanoma Media Finland announced on 4 September that refusing certain cookies will require a paid subscription from October, with the fee set against the advertising revenue foregone, citing an August precedent from Finland’s Supreme Administrative Court. CMA Media launched the free RMC+ platform on 1 September with nearly 10,000 hours of content, while 93% of BFMTV and RMC advertising revenue still comes from linear television.

Each of these is a decision about where a publisher intends to sit on the barbell. Ouest-France is defending scale in a regional franchise by cutting fixed cost. Sanoma is pricing consent, which converts a compliance obligation into a revenue line and tests how much a Finnish reader will pay to refuse tracking. CMA Media is buying optionality in streaming while acknowledging that its cash still comes from broadcast advertising.

The barbell hypothesis predicts that the thin part of the bar is the general-interest, mid-scale, advertising-funded product with no distinctive rights position. What these five signals show is that publishers are aware of it and are choosing an end deliberately. The decisive test will be whether the products at each end can be financed from the same balance sheet during the transition.

Reflections

If a reader can pay to refuse tracking, what does that price reveal about the real value of consent in your market?

Which position on the barbell is your organisation actually financing today, judged by where new investment goes?

V. On rights, regulation and case law

Three signals in which courts, competition authorities and legislators define the terms on which content and synthetic output circulate.

27.8 — Munich holds that memorisation by a model is reproduction

On 31 July the 42nd Civil Chamber of the Regional Court of Munich I ruled largely in favour of GEMA against Suno in case 42 O 763/25, granting injunctive relief, information and damages claims concerning six works including “Atemlos durch die Nacht”, “Forever Young” and “Daddy Cool”. The court found that the v3.5 and v4 models had memorised reproducible parts of works obtained from YouTube through stream-ripping that circumvented its rolling cipher, held that the American training reproductions were outside the exception invoked in defence under 17 U.S.C. §107, and held that the reproductions inside the models and in their outputs in Germany fell outside the text-and-data-mining exception in §44b UrhG. GEMA, which describes the judgment as its second court victory against an AI provider after the November 2025 OpenAI ruling now on appeal, states that licences must be acquired; the judgment is open to appeal. The French-language analysis by Marie-Avril Roux Steinkuehler stresses the same distinction the court drew, between analysing data and storing recognisable versions of protected works.

Two American filings frame the contrast. The United States government filed a brief on 1 September supporting OpenAI in the New York Times litigation, arguing that AI training is extraordinarily transformative and that the country has a strong interest in rejecting the newspapers’ position; a brief carries advisory weight, the first two judges to address the question reached diverging conclusions, and the outcome remains open. Four days later, the Seattle Times and Newsday sued OpenAI and Microsoft in the Southern District of New York, alleging that paywalled journalism was copied into training datasets and seeking destruction of the datasets and models involved. Stibbe counts more than 70 active or recently resolved AI copyright actions, notes that Article 4 CDSM allows a machine-readable reservation of rights, and flags the Advocate General’s opinion in Like Company v Google, C-250/25, expected this month.

The European position is being established through copyright reasoning rather than through imported doctrine, and the Munich chamber has located the decisive question with precision: the boundary is crossed where analysis becomes storage of recognisable protected material. Claims of this kind establish the legal limits; licensing negotiations establish the workable price; a business model emerges from the combination of the two. Both halves are necessary, and neither substitutes for the other.

Reflections

How should a rights holder express a machine-readable reservation of rights so that it is enforceable and auditable rather than merely declared?

What would a licence look like that priced memorisation separately from retrieval, and who would verify the difference?

27.9 — Regulators move from principles to instruments

EU antitrust regulators sent publishers a questionnaire in July, with an 28 August deadline, on Google’s proposal to let them opt out of AI search without losing traditional search ranking; the opt-out was announced in June, completed globally on 31 August after a UK order, and the investigation could lead to a further fine on top of more than 10 billion euro already imposed. On 1 September the European Commission designated ChatGPT as a Very Large Online Search Engine under the Digital Services Act, alongside Reddit and Roblox as platforms, giving them until January 2027 to assess and mitigate systemic risks. South Korea’s media commission has requested 34.6 billion won for 2027, including 10 billion won to train AI on broadcasters’ stock footage and funding for a fact-checking transparency centre.

Each instrument reveals a different signal about the same market. The competition file asks whether an opt-out can be genuine when the same company controls the ranking that publishers depend on. The DSA designation of a conversational service as a search engine confirms that answer interfaces have become the gateway, with the systemic-risk obligations that follow.

Taken together, this is regulation shifting from statements of principle to countable instruments: a machine-readable opt-out, and a designated gateway with audit duties, Instruments create evidence, and evidence is what makes a licensing market function.

Reflections

Under what conditions would you use an AI search opt-out, and what would you need to see in the traffic data before trusting it?

What systemic risks can be created by a chatbot that qualifies as a search engine?

27.10 — Rules for synthetic content and information conduct take shape

Brazil’s top electoral court defined a deepfake, by a 5-2 vote, as synthetic content produced or manipulated through AI or equivalent technology that is realistic enough to create, reproduce or alter the image, voice or speech, setting new standards for identifying AI-generated deepfakes in political ads, ahead of a first presidential round on 4 October. In China, regulations on multichannel distribution services took effect on 1 September, requiring MCN agencies to register, appoint a content-management director, maintain a permanent monitoring team and terminate creators who breach the rules, in a sector of more than 40,000 agencies and roughly 100,000 signed influencers; the reporting by the Croatian agency HINA, notes an estimated 100 billion yuan market. The same authorities reported on 3 September the removal of more than 5.61 million items and action against over 49,000 accounts in a campaign against AI misuse, building on labelling rules in force since September 2025. Mexico’s draft guidelines on audience rights would require broadcasters to separate news from opinion and appoint ombudspersons, with fines up to 1% of annual revenue, and have drawn criticism from CIRT and from Reporters Without Borders over ambiguity and anonymous complaints.

These four measures are attempts to regulate the new reality through different constitutional tradition based on the reliability of what an audience sees. A court definition of synthetic content is narrow and enforceable in an electoral context. An administrative registration duty for intermediaries is broad and gives the state leverage over the entire creator economy. A rights-of-reply regime for broadcasters addresses accountability and invites the objection that the regulator ends up assessing editorial judgement.

The comparison matters for European practice because the technology moves faster than the safeguards. A workable definition of manipulated media, tied to a specific harm and a specific procedure, is more useful to newsrooms than a general prohibition, and considerably harder to write.

Reflections

Which legal definition of synthetic content could a newsroom apply in the hour before publication rather than in litigation afterwards?

Where does an audience-rights regime stop protecting audiences and start shaping editorial judgement?

VI. On the economics of attention and the cost of verification

Two signals on how the open web now generates revenue, and on what it costs to keep the record accurate.

27.11 — Machine visits grow while human sessions decline

Digiday reports that the open web is separating into human audiences and AI crawlers, citing People Inc.’s 22% year-on-year fall in core sessions and 40% decline in Google search traffic against flat session-based revenue, 16% growth in non-session revenue and a rise from 39% to 43% of digital revenue; chief innovation officer Jon Roberts states that the open web is getting smaller, while Forbes describes AI licensing models as nascent. Press Gazette reports 51Degrees analysis of three billion visits showing that AmazonBot accounted for more than 4% of all visits and Meta-External-Agent almost 3%, that AI bots reached 13% of web sessions in May 2026 against 8% a year earlier, and that publishers block only around 21% of AI scrapers in robots.txt. The Journalism Relay Project and the Media Leadership Think Tank found in their study of 263 South African news sites that only 30.4% block at least one AI crawler, that leading sites lost about a fifth of daily page views in a year, and that chatbots referred under two million visits, less than 1% of 256 million chatbot visits recorded.

Machine consumption is rising towards a significant share of infrastructure load while contributing almost nothing in referral traffic, which means the cost of serving content is migrating away from the revenue it once produced. The South African figures show how this creates discrimination: blocking requires engineering capacity, so smaller publishers remain the most accessible and the least compensated. Michael Markovitz of the Media Leadership Think Tank describes robots.txt accurately as a yes-or-no gate rather than an enforcement tool.

This is why the SPUR standards work in signal 27.1 has commercial substance. A gate answers one question; a meter answers the questions that produce an invoice, namely which content, how often, for what purpose and at what value.

Reflections

What share of your infrastructure cost is now attributable to machine consumption, and who inside the organisation owns that number?

What answers does the standard meter has to come up with to mesure value creation?

27.12 — Verification becomes the visible product

Reuters has published an account of how its global visual verification team reviews hundreds of images daily and publishes perhaps a dozen, interviewing original photographers, examining metadata, comparing shadow direction with satellite, archive and street-view imagery, and scanning material with several detection tools that the agency states are imperfect. The Society of Professional Journalists is revising its Code of Ethics for the first time in twelve years, with a 967-word draft going to a vote in October that addresses AI, verification, fabricated images and creator journalists, and with vice chairman Chris Roberts stating that journalists remain ethically responsible for their work regardless of the tools used. The demand side of verification is documented in Meta’s disclosure that it disrupted an Iran-based network of 35 accounts using AI-generated posts followed by about 79,400 Instagram accounts, in NPR and NewsGuard’s finding that chatbots debunked false narratives about three-quarters of the time while AI summaries performed worse, and in Queensland University of Technology research showing that 12.5% of analysed AI Overviews on 15-minute cities presented false claims as one side of a legitimate debate.

The failures now reach the institutions that consume journalism. Guardian Australia found at least 39 parliamentary submissions containing apparently fabricated references, with more than 100 papers carrying ChatGPT URL tags, including misattributed work by named academics. Press Gazette established that four prolific technology and finance freelancers could not be verified as real people, leading Hacker Noon and Information Age to remove articles, in a pattern linked to promotional campaigns for cryptocurrency projects. Bernama’s chief executive used the Asia-Pacific Media Forum in Shenzhen to call for cross-border fact-checking and shared principles for responsible AI use, and a research consortium is surveying journalists in the United Kingdom, Brazil, Canada, the Netherlands and South Africa on the professional qualities that define the craft.

Verification has moved from an internal cost to an externally priced capability. Reuters publishes its method because the method is the product, and an agency that can demonstrate provenance holds an asset that scales with the volume of synthetic material in circulation. That is the strength of the quality newsroom as an institution: a repeatable procedure, documented and auditable, rather than the reputation of any individual byline.

Reflections

What would you charge for verification if it were sold as a service rather than absorbed as a cost of publication?

Which parts of your verification method can you publish without weakening it, and what does publishing them achieve?

Conclusion

The barbell hypothesis explains most of what this week’s signals contain, and it leaves one question open. If the market divides into boutique journalism and large efficient services, the position worth holding is the bar itself: the layer through which verified content, its rights and its usage records pass between producers and machines. Whoever specifies that layer determines how value is distributed across everything attached to it.

This is why the EANA decision to join a standards coalition belongs at the front of this edition. Agencies occupy that middle position by construction, since they sell structured rights and verified material into every other part of the system. CNA pricing machine access, AFP preserving editorial metadata inside a production system, Ritzau guaranteeing the origin of an official message and Reuters publishing its verification method are all versions of the same commercial claim: the scarce good is proven provenance, and it costs money to produce.

The legal picture supports that claim from two directions. Munich has established that storing recognisable protected works inside a model is reproduction under German and American analysis alike, while Brussels has designated a conversational service as a gateway with systemic-risk duties. Claims define the boundary, negotiations set the price, and a functioning market requires both. The American filings this week are a reminder that the boundary is contested rather than settled, and that European rights holders should build their position on copyright reasoning and on machine-readable reservations they can actually enforce.

For the operators, the practical agenda is narrow enough to act on. Know what proportion of your infrastructure serves machines and what that costs. Express your rights in a form a system can read, and your terms in a form an agent can transact against. Decide which capability in your organisation is genuinely scarce, and price it before someone else standardises it for you.

Markus Knall’s closing observation deserves the last word, because it points past the tooling to the demand side. If every process and every piece of content becomes AI-optimised, the remaining reason for a reader to choose your journalism is the confidence that what you publish is true and that something new has been discovered. That confidence is the one asset in this market that cannot be generated, only earned and then documented.

This week’s links are about those three fronts. Copyright litigation and licensing keep defining what training data is worth. Public authorities in Europe and Asia move from principles to procedures, on cybersecurity testing, on takedown rules and on newsroom AI policies. Press agencies and large publishers turn trust itself into a product line, from Reuters on Snowflake to AFP’s factchecking service and Thomson Reuters’ own frontier model.

The pattern is direct. Regulators, audiences and enterprise buyers now ask journalism to function as reliable infrastructure in an information system that industrialises around it. The business question is whether that infrastructure can earn a price before the audience relationship becomes intermediated.

This image proposes a working hypothesis for the whole dispatch. A lighthouse in the middle of the Cyber Territories metropolis, a glass tower where journalists and AI systems work together, casts one defined beam of verified information across a city crowded with deepfakes, bot networks and manipulated imagery. Below the tower, secured data channels reach paying users, businesses and machines through subscriptions, digital products, B2B data services, API access and licensing partnerships. Quality journalism as reliable infrastructure in an industrialised information environment, priced and paid for through a portfolio of revenue lines. (Synthetic Image)

I. On law, licensing and the price of training data

One tension marks this chapter, the gap between how the United States and the European Union treat text and data mining, and how publishers learn to monetise the same corpus on both sides.

26.1 — US copyright fights over AI training remain unresolved, and do not translate to Europe

TechCrunch’s analysis of the ongoing US litigation over training AI models on copyrighted books sets out how uncertain the legal picture remains, even after the first round of district-court decisions. The piece frames the debate around the American fair use doctrine and the question of whether ingestion for training qualifies as transformative enough.

European readers should treat this as a United States conversation. The fair use doctrine does not exist in the European Union; the relevant framework is copyright law, and the moral and related rights that surround them. Presenting US outcomes as a general answer, misreads both the litigation record and the applicable European law.

Even inside the American system, extending fair use to full-corpus ingestion for commercial foundation models remains a considerable stretch. The cases that have gone furthest still involve narrow factual settings, and appellate review has barely begun. European rights holders should keep negotiating licences on European terms and treat US commentary as background rather than as precedent.

Reflections How should European publishers frame licensing offers when their US counterparts point to fair use style arguments as an alternative to paying? What safeguards will keep copyright enforceable when AI companies train models outside the EU and then deploy them inside it?

26.2 — USA Today reformats content for AI licensing, and Reuters joins Snowflake

Two commercial signals emerged in the same week. Digiday reports that USA Today Network reformats its content specifically to attract more AI licensing deals, restructuring articles and metadata so that AI buyers can ingest and cite them more easily. In parallel, Reuters announced that it has joined the Snowflake Marketplace to offer AI-ready, trusted news content to enterprise customers who want licensed, verifiable inputs into their own models and analytics pipelines.

Both moves recognise that the buyer changes. Publishers now package content for machines and for the teams that build with them, alongside the traditional B2C audience. The value proposition rests on provenance, structure and legal cleanliness, exactly the attributes that general web collection cannot guarantee.

Here lawsuits and licensing meet as two halves of one process. Litigation establishes the legal boundary of unlicensed training; marketplace deals establish what a workable price looks like once publishers enforce that boundary. Publishers who invest in structured, machine-readable formats today will hold more leverage in the next round of negotiations.

Reflections Which parts of a news archive create the most licensing value once publishers structure them properly? How can smaller European publishers pool their content into offers that enterprise AI buyers can read at scale, without ceding control over pricing?

26.3 — Thomson Reuters launches its own frontier model, Australia debates platform pay-up rules

Thomson Reuters announced that it leverages its world-class data assets to launch its own frontier model, aimed primarily at legal, tax and professional research use cases. The company bets that curated, high-value proprietary data plus a domain-specific model will beat generic frontier systems on the tasks its clients actually pay for.

At the political end of the same dynamic, AAP reports that Australia’s major parties agree on new pay-up rules for social media platforms, designed to secure payments to publishers whose content circulates on those services. This iteration extends the model that began with the News Media Bargaining Code.

Together these moves describe the same strategic logic from two directions. Owners of scarce, verified content move up the stack, either by building their own AI products or by requiring platforms to pay for the material that already fuels theirs. Quality journalism functions as a specialised input that carries a real cost of production, and European publishers should treat it that way.

Reflections When a large publisher operates its own frontier model, how should it price access for smaller newsrooms that could become both customers and competitors? Which regulatory design keeps platform pay-up rules from turning into a permanent subsidy that dulls the incentive to innovate on the publisher side?

II. On sovereignty, security and adversarial information

The state moves back into the AI and information stack, in Europe as well as in Washington and Beijing. Security testing, model oversight and influence operations converge into one file.

26.4 — France uses AI to test its own cybersecurity, and Brussels probes the labs

Reuters reports that France will use AI tools to test its cybersecurity after a hack at the tax agency, a project framed explicitly around national resilience and the country’s ambition to keep sovereign AI capacity, including Mistral, in the loop. In the same week, Euractiv obtained the letter in which the European Commission ordered leading AI labs to detail their security practices, covering red-teaming, model evaluations and incident reporting.

Together these two moves show a European stack forming around AI sovereignty. The debate now extends beyond what the United States and China do, with the EU asserting jurisdiction over how frontier models get secured, tested and audited on European infrastructure.

For newsrooms, sovereignty carries operational weight. It determines which models editors can integrate into editorial workflows without breaching data-protection or contractual obligations, and which vendors will certify that their systems meet the emerging European baseline.

Reflections How should a European newsroom weigh the operational advantages of frontier US models against the compliance and sovereignty case for European alternatives? Which security disclosures from AI labs genuinely help editors and CISOs, and which risk becoming compliance theatre?

26.5 — OpenAI disrupts a Russian ChatGPT campaign, and Yonhap sets the standard for how to report a platform-led disclosure

CNBC reports that OpenAI disrupted a Russian influence operation that used ChatGPT to generate propaganda content, part of the company’s periodic threat reports on state-linked misuse. Almost simultaneously, X said it identified a Chinese bot farm targeting the debate around AI data centres. The underlying story comes from Yonhap, the South Korean press agency, and the choice of source matters. Yonhap does exactly what a press agency should do with a platform-issued announcement of this kind: it reports the disclosure as a claim by X, attributes each element to the platform, adds independent context from South Korean cybersecurity researchers and public-sector sources, and refrains from treating the platform’s own communication as a verified finding. That editorial posture, precise attribution and independent framing, is the template other newsrooms should follow when they cover the same file.

Two things stand out in the underlying substance. First, the adversarial use of AI to generate content and to amplify it has become routine, and the actors behind it match those long identified in threat-intelligence work. Second, the disclosures come from the platforms themselves, which puts newsrooms in the position of reporting on evidence they cannot fully verify without cooperation from those same platforms.

Coverage should reflect that dependency, as Yonhap’s does. When a platform publishes a takedown announcement, editors should report it as a claim, seek independent confirmation where possible, and place it in the wider pattern of state-linked influence operations that intelligence services have documented for years.

Reflections What editorial standards should govern the reporting of platform-issued takedown claims, so that they inform readers without becoming free public relations? How can independent researchers get sustained access to the data needed to verify or challenge these announcements over time?

III. On verification, trust and editorial guidelines

Trust rebuilds through specific practices, most of them unglamorous. Factchecking desks, visual verification protocols and formal AI use policies do the day-to-day work of that reconstruction.

26.6 — AFP’s factchecking service and Reuters on visual verification

AFP’s dedicated factchecking desk on AI has grown into one of the most consistent public references on synthetic media, coordinated influence and manipulated imagery. A recent example illustrates the method. AFP debunked an AI-generated video that circulated as real footage of the Nepal-Tibet floods, ran it through Google’s SynthID detector, and documented the visual anomalies typical of AI generation. AFP’s own factchecking manual codifies the underlying approach: transparency about sources, prioritisation of harmful and viral claims, and detailed public explanations of the method behind each verdict. In parallel, Reuters published a piece explaining what visual verification is and why it practises it, describing the systematic checks its picture and video desks apply before publication.

These pieces also position both agencies commercially. A press agency that can demonstrate a documented, repeatable verification process turns that process into a market asset. Enterprise clients, platforms and other newsrooms have a reason to buy or licence access to it, precisely because independent verification at scale carries a high cost of production.

The wider pattern shows verification becoming a specialised industrial function, tightly coupled to press agencies as critical information infrastructure. Public broadcasters and quality news publishers can rely on it, integrate it and cite it, provided the underlying business model of the factchecking desks remains viable.

Reflections Which combination of subscription, licensing and public funding will keep large-scale factchecking desks sustainable over the next decade? How should a newsroom credit and cite external factchecks in its own reporting, so that readers understand where the verification actually happened?

26.7 — Editorial AI guidelines from Amman, EFE and CMN, and NYT tests AI search summaries

Three newsrooms formalised their AI policies in the same window. The Community Media Network in Amman amended its editorial policy to establish professional guidelines on AI use. Spain’s Agencia EFE published a public guide to AI use inside the agency. And Vietnam has officially adopted a formal process for handling fake and false news, which gives the government a codified procedure that will shape how domestic media operate.

On the product side, Semafor reports that The New York Times tests AI-generated search summaries on its own site, an experiment that continues alongside its ongoing litigation against OpenAI and Microsoft.

The common thread is procedural. Whether the driver is professional ethics in Amman and Madrid, state control in Hanoi, or product experimentation in New York, newsrooms and regulators move from principles to written procedures. That formalisation makes AI use auditable, both internally and for the audience.

Reflections Where should the line settle between disclosing AI use to readers in detail and overwhelming them with technical caveats on every article? How can a written newsroom AI policy stay useful when the underlying tools change every few months?

IV. On the shape of the newsroom, and its audience

The final chapter examines the newsroom itself, its people, its automation choices and the channels through which its work reaches readers.

26.8 — Automation inside the newsroom, from AP school boards to Sport in Spain, Nine and LatAm

Four pieces describe the operational reality. The Associated Press launched AP School Board Search, a tool built to transform local coverage of thousands of school boards by making local records searchable and reportable at scale. In Spain, WAN-IFRA describes how the sports daily Sport explores AI for its digital newsroom, integrating generative tools into production while keeping editorial control. In Australia, The Guardian reports that Nine Entertainment signs AI deals and cuts costs across the network. And Latam Journalism Review covers Google and Futura Lab launching an AI programme for newsrooms in Chile, Ecuador, Paraguay, Peru and Uruguay, with parallel growth support for eight South African news creators through Google’s Growth Lab and the AI for Media Network hackathon in 2026.

A separate signal completes the picture. The Blaze profiles a startup that automates investigative journalism and asks whether that automation serves the public interest at all. The answer depends on whether the automation extends what human investigators can do, or replaces the judgement calls that make an investigation defensible in court and in public.

Two patterns emerge. First, newsrooms deploy generative tools to unlock coverage that previously remained uneconomic, from thousands of school boards in the United States to sport verticals in Madrid. Second, a substantial share of the training, funding and tooling for that transition comes from a single dominant platform, which raises a question about the resilience of any newsroom that depends on it.

Reflections Which automation choices genuinely expand a newsroom’s editorial capacity, and which merely reduce headcount without adding new coverage? How should quality newsrooms structure their partnerships with dominant platforms so that they gain capability without ceding strategic autonomy?

26.9 — AI reshapes distribution, from Google’s AI search to Parag Agrawal’s future web, AFP–Dalet, and the newsletter question

TechCrunch, citing new data, reports that Google’s AI search rapidly becomes the default, which continues the shift of referral traffic away from ranked links and towards summarised answers. Parag Agrawal offers a longer view on AI search and the future web, describing a web that reorganises itself around retrieval by machines as much as by humans. AFP and Dalet have announced an integration that brings trusted content directly into newsroom production tools, a business-to-business distribution move rather than a consumer one. INMA asks whether newsletters will survive the AI assault on email, as inbox summarisation and AI-native readers start to intermediate the last direct channel publishers own. FT Strategies interviews Angela Mackay on what comes beyond reader revenue, pointing towards a mix of subscriptions, B2B products and licensing.

The distribution question has become structural. AI summaries reshape consumer traffic through general search; B2B channels and content marketplaces expand enterprise distribution; and AI assistants themselves mediate the direct channel of email.

Newsroom leaders should plan for a portfolio of distribution surfaces at once, general search, AI answer engines, licensed enterprise feeds, direct subscriptions, newsletters and audio, on the understanding that no single surface will remain stable enough on its own.

Reflections If a majority of consumer readers first encounter your journalism inside an AI answer rather than on your site, which parts of your business model still work? How should a quality newsroom decide which distribution surfaces to invest in, and which to accept as commodity channels?

Conclusion

Dispatch 26 traces the outline of journalism as infrastructure. Publishers, regulators and platforms rebuild the newsroom into something that resembles a public utility for verified information, with private capital, public interest and platform power pulling on it from three sides at once.

The legal front now has visible contours. Europe operates on copyright and related rights, with licensing agreements as the working currency. The United States operates on fair use, with unresolved litigation and settlements that have not yet established a stable price. European publishers should keep the two regimes distinct and negotiate on their own legal ground.

The security and sovereignty front moves fast. France testing its own systems with AI, the Commission auditing frontier labs, OpenAI disrupting Russian influence campaigns, X flagging a Chinese bot farm and Yonhap covering it with the professional caution the file deserves: these pieces belong to the same picture, in which governments and the industry treat the information layer as critical infrastructure.

The editorial front reaches the reader directly. AFP’s factchecking desk and Reuters’ visual verification protocols show what industrial-grade trust looks like when agencies build it properly. AI use policies at CMN, EFE and inside Vietnam’s official framework show how quickly the practices become codified. The New York Times testing AI summaries on its own pages shows that a single organisation can litigate against generative players and integrate their techniques into its own product at the same time.

The strategic question for the next quarter concerns which of those revenue lines a given newsroom can defend, and which it will need to build from scratch. The answer will differ by market, by size and by the strength of the underlying archive. What all cases share is that the audience relationship, the licensing relationship and the platform relationship now belong on one portfolio and on one strategic map.

Seventy years after a handful of European news agencies signed a founding text in Strasbourg, their successors meet again as EANA, the European Alliance of News Agencies representing 33 member organisations across 33 countries and more than 670 million people. The anniversary statement names the tensions that now define our week: copyright and content provenance, AI-driven disinformation, and threats to the safety of journalists.

Around that anchor, this week’s material sketches a wider picture. AI systems keep expanding into the writing, distribution, and verification of news, while regulators, publishers, unions, and readers each search for a viable equilibrium.

Three threads run through the signals. First, the industrial production of AI text and imagery is now measurable, and the tools to detect and label it are catching up. Second, the economic relationship between publishers and AI platforms is moving from anecdote to data, with early evidence that licensing deals shape citation shares. Third, the human premium, in journalism as elsewhere, is becoming a strategic argument, and one that press agencies as critical information infrastructure are well placed to make.

The Strasbourg text of 1956 spoke of cooperation between independent agencies. In 2026 that same principle, applied to AI governance and to the defence of verified information, is what gives the anniversary its practical meaning.

I. On press agencies and the anniversary that frames the week

One event sets the register for everything that follows.

25.1 — EANA marks seventy years and names the agenda for the next decade

The European Alliance of News Agencies celebrated its 70th anniversary on 21 August 2026, recalling its founding in Strasbourg in August 1956 and taking stock of a network that now spans 33 member organisations serving more than 670 million people. President Stefano De Alessandri stated that “verified information serves the public interest, regardless of national borders” and that cooperation is “the foundation on which independent journalism survives”. Secretary General Alexandru Ion Giboi added that “our members are what makes our organisation strong”, while AGERPRES Director General Claudia Nicolae described the Alliance’s legacy as one of “solidarity, cooperation, and a common voice”.

The statement identifies three pressures: copyright and content provenance, AI-driven disinformation, and threats to journalist safety. It also confirms EANA’s position on generative AI, summarised in the principle of “no consent, no content” for the use of news agency work in training. This is a clear European frame for the licensing debate, and it belongs alongside the copyright directive as part of the equilibrium that publishers, agencies, and AI developers are now negotiating.

For CEOs and policymakers, the anniversary is more than a commemoration. It positions news agencies as critical information infrastructure, distinct from news publishers and public broadcasters, and it names cooperation, innovation, and high journalistic standards, points highlighted by board member Aimilios Perdikaris of ANA-MPA as the conditions for economically sustainable journalism.

Reflections If “no consent, no content” becomes the European default for AI training on agency material, which licensing models can carry that principle into workable prices? How can a seventy-year-old alliance translate its heritage of cooperation into governance capacity for AI, provenance, and safety in the next decade?

II. On the industrialisation of AI text and the search for provenance

Detection, labelling, and measurement of machine writing all moved forward this week.

25.2 — Anthropic watermarks Claude, and the debate over what writing is

Anthropic has begun adding an invisible watermark to text generated by Claude, explained on Nieman Lab as a system that deliberately selects slightly less-likely words in detectable patterns, for example “guava” instead of “mango”. The Guardian reports that the change responds to an EU regulation requiring watermarking of AI-generated text from December, with the pattern detectable by Anthropic and by holders of a decoding key. Critics such as John Gruber argue that the method is “a perverse adulteration of what it means to write”, because it constrains the model’s choice of the most precise word.

The technical detail matters for editorial policy. Steven Murdoch of UCL considers the perceived quality loss likely to be small, since language models already rely on controlled randomness to avoid repetition and loops. At the same time, watermarking is one of the few tools that can address model collapse and provide a signal for detection at scale.

For newsrooms, the practical question is whether watermarking becomes a standard input for provenance workflows alongside SynthID and C2PA, or whether it remains a compliance layer visible only to the model provider. The European regulator has set a direction; publishers and agencies now need to decide how detection results are used inside editorial checks, and how they are disclosed to readers.

Reflections If watermarks become mandatory for large models operating in the EU, what verification duties should editors assume before they publish or quote AI-assisted text? How should a publisher weigh a marginal loss in prose quality against a measurable gain in traceability for the information ecosystem?

25.3 — Pew measures how much of the web is now written with AI

A Pew Research Center analysis of almost half a million English-language webpages estimates that one in ten pages collected in July 2026 show significant signs of AI authorship, rising to more than one third among pages published after ChatGPT’s release. The share reaches roughly one in ten on .com pages, against 4.6% on .org and around one percent on .edu and .gov. Compared with 2023, em dashes appear about twice as often, Oxford commas increase by 63%, AI-associated words more than double, and negative parallelism nearly triples.

The figures give a first stable baseline for a phenomenon that has so far been discussed in impressions. They also confirm the stylistic signatures that professional editors already recognise, and they suggest that institutional and academic domains have adopted AI writing more slowly than commercial ones.

The New Yorker’s essay on “the prehistory of AI slop” places the current wave in a longer history of mechanised writing, from eighteenth-century letter-writing manuals to Wycliffe A. Hill’s 1931 “Plot Robot”. Read together, the two pieces suggest that formulaic writing has a long history, and that today’s change lies in scale and cost, which have shifted by orders of magnitude and turned the current cycle into a governance problem for information ecosystems.

Reflections When one in three post-2022 web pages carries signs of AI authorship, what standard of disclosure should quality newsrooms adopt to remain distinguishable? How can historical perspective on formulaic writing help editors set proportionate rules for AI use, rather than reactive ones?

25.4 — When AI writes shorter, not just faster: the FT column that crossed a line

Poynter reports that the Financial Times appended a note to a column by Harvard professor Ricardo Hausmann, disclosing that AI had been used to condense a longer draft before submission, in breach of the FT’s editorial code, which prohibits AI in the writing process. The FT’s principles allow AI for research, transcription, and data analysis with human oversight, while insisting that journalism must be “reported, written and created by our journalists and editors”.

The case sharpens the question posed by the FT itself: when does an editing tool become a writing tool? Condensation reshapes emphasis and argument, and it is difficult to argue that a machine-condensed column is still fully authored by the byline it carries.

For editors, the practical takeaway is that AI policies need to reach outside contributors, including columnists and guest authors, and to specify which editorial acts remain reserved to humans. The line is not obvious, and every organisation will draw it in a slightly different place; making the rule explicit is the minimum condition for trust.

Reflections Which editorial acts, from copy-editing to structural condensation, should remain reserved to human authors even when disclosed? How should a newsroom respond to a contributor who has crossed that line, once disclosure has been made?

III. On synthetic content, information disorder, and elections

Where AI content is unlabelled or manipulated, the effects on public information are already visible.

25.5 — Deepfakes are the headline; election infrastructure is the risk

Rest of World argues that the real danger of AI in elections lies in the systems embedded in voter registers, biometric verification, and results software, rather than in the deepfakes that dominate coverage. The piece describes India’s 2015 programme linking election records to Aadhaar, in which about 5.5 million voters were removed in two states before the Supreme Court intervened, with a later algorithmic failure rate reported at 93%. It also flags growing reliance on private vendors across Africa, Asia, and Latin America for election technology, raising questions about foreign data transfers, unauditable proprietary systems, and chatbot misinformation from consumer AI products.

Poynter’s account of Median Strategies’ fake polls, which briefly influenced coverage of Los Angeles and Wisconsin races before real results contradicted them, makes the same point at a smaller scale. Poynter’s Kelly McBride calls reliance on an unknown pollster without published methodology “inexcusable” and reminds editors that polls are estimates, not predictions.

For democratic institutions and news organisations, the combined message is that AI-related electoral risk is a supply-chain problem. Auditability of vendors, transparency of methods, and disciplined use of poll data belong on the same governance list as deepfake response protocols.

Reflections Which parts of election infrastructure should be legally required to disclose the AI components they use, and to whom? How should newsrooms verify pollsters and datasets before amplifying figures that can affect turnout and expectations?

25.6 — Fake newsrooms, fake pitches, and the effort to re-humanise the process

Interlochen Public Radio reports that Prism News has shut down its network of more than 200 AI-generated local news and hobby sites, including titles that posed as local newsrooms under fabricated bylines such as “Dr. Elena Rodriguez”. The Israel-based startup had presented itself as an answer to news deserts but produced no original coverage, aggregating and slightly altering material from real outlets. In parallel, Nieman Lab describes The Food Section’s decision to accept story pitches only by phone from September, after founder Hanna Raskin estimated that a quarter of recent pitches were “wholly AI”.

Two responses, at opposite ends of the market, converge on the same insight: when synthetic content becomes cheap enough to industrialise, verification and human contact become scarce goods. Closing a 200-site network and opening a pitch hotline are complementary moves, one enforcing legality and reputation, the other rebuilding editorial trust at the front door.

The Verge’s report on Reddit’s experiment turning text posts into AI-voiced short videos, labelled “Real conversation voiced by AI”, shows the same logic on the distribution side. As synthetic formats proliferate, clarity about what is machine-produced and what is not becomes an editorial obligation, and a competitive asset.

Reflections What legal and platform standards should govern networks of AI-generated sites that impersonate local journalism? How can quality newsrooms turn “human by design” workflows, such as phone-only pitches, into a durable brand signal?

IV. On tools, skills, and the human premium

New tools are helping newsrooms verify, and new skill requirements are reshaping the labour market.

25.7 — Google’s Backstory and the shape of AI-assisted verification

Nieman Lab describes Google DeepMind’s Backstory, an experimental tool that helps fact-checkers determine whether images are AI-generated or manipulated and where they have appeared online. Built on Gemini models and AI agents, it checks SynthID watermarks and C2PA credentials, produces a cited report, and logs its steps. India Today team leader Bal Krishna says the tool has “no doubt” sped up daily work by combining tasks that once required “five different tools, five different logins”.

The design point is important: Backstory is presented as a first pass, not a source of citation. India Today does not cite it directly and requires reporters to verify findings manually. Digital literacy expert Mike Caulfield estimates it can reduce a 50-minute verification process to three minutes for indexed material, while noting that images from TikTok, WhatsApp, or Signal remain outside its reach.

For editors, the implication is a workflow question. Verification will increasingly rely on assisted tools whose outputs must themselves be validated, and provenance signals such as SynthID and C2PA will only be useful if editorial systems can read and record them at scale.

Reflections Which verification steps should remain manual as assisted tools such as Backstory become standard first-pass instruments? How should newsrooms disclose to readers the role of AI in their fact-checking, without weakening confidence in the finding?

25.8 — Recruiters raise the AI skills bar, and the “human premium” grows

Journalism.co.uk reports that the NCTJ has found news employers increasingly demanding advanced AI skills from journalism graduates, including critical prompting, fact-checking, context awareness, ethical judgement, and understanding of bias and hallucination. Its research covered 14 UK news organisations and shows that every one has adopted AI to some degree. In parallel, Forbes summarises reports from LinkedIn, Visier, and the World Economic Forum arguing that AI is creating a “human premium” around judgement, communication, curiosity, critical thinking, and interpersonal intelligence.

Read together, the two pieces describe a bifurcated skills market. Technical AI literacy is now baseline, while the returns to distinctively human capabilities, applied to editorial judgement and audience relationships, are increasing.

For news agencies and publishers, this argues for two coordinated investments: structured AI training for staff at every level, and explicit recognition of the editorial and interpersonal work that AI cannot substitute. Boyd Taylor Coolman’s essay in America Magazine on Pope Leo XIV’s warning, in “Magnifica Humanitas”, that AI can “extinguish the desire to ask questions” offers a philosophical counterpart. Coolman’s recovery of a medieval account of intellectual virtue is a reminder that reading, thinking, and questioning are worthwhile in themselves, a point with practical consequences for how newsrooms design work.

Reflections Which AI skills should be included in the core curriculum for every journalist, and which should remain specialist tracks? How should organisations recognise and reward the human capabilities that AI cannot replace, without turning them into a nostalgic argument?

25.9 — Cooperation and skills at continental scale

Two international items complete the skills picture. Africanews reports on the 7th Forum on China-Africa Media Cooperation in Beijing, where African Union of Broadcasting CEO Gregoire Ndjaka stressed that Africa must “develop the skills to understand, adapt and use” AI technologies rather than simply consume them. Mediabites summarises WAN-IFRA CEO Stig Kirk Ørskov’s first-summer assessment of global journalism, in which he warns that AI-powered search and chatbots increasingly answer directly and remove publishers from the referral chain, calling the current platform arrangement “not a sustainable foundation for a healthy information ecosystem”.

The two views reinforce each other. Skills, cooperation, and the defence of professional journalism are converging into a single agenda, and it is one that international alliances, from EANA to WAN-IFRA to the AUB, are increasingly expected to lead.

For strategy leaders, the practical question is how to align internal training programmes, external partnerships, and public advocacy so that the “human premium” is not only an HR argument but a distinctive market position.

Reflections Which cooperative structures are best placed to build AI skills across small and mid-sized newsrooms in Europe and beyond? How should industry alliances translate skills investment into pricing power when negotiating with AI platforms?

V. On law, licensing, and the economics of AI in news

Rules and money are moving together, and the shape of a workable equilibrium is starting to emerge.

25.10 — The EU AI Act as an editorial reference beyond Europe

ISS Africa argues that South African newsrooms and legacy media should treat the EU AI Act as a reference for transparency, accountability, and information integrity in an African context. Konrad-Adenauer-Stiftung Media Programme Director Hendrik Sittig is quoted as saying that the Act permits newsroom AI use while requiring “greater transparency, clear labelling and accountable editorial oversight”. A CINIA study cited in the piece finds that South African AI adoption often depends on individual employee interest, while few newsrooms have clear policies and some journalists reportedly use AI “secretly”.

The article documents the risks that result, including fake citations in a draft National AI Policy and alleged non-existent case law in a Johannesburg acting judge’s ruling. It also references the Press Council of South Africa’s guidance that AI-generated material be checked by “human eyes and hands”, fact-checked, and transparently identified.

For European agencies and publishers, the wider point is that the AI Act is becoming an editorial and reputational reference well beyond the EU’s jurisdiction. Consistent labelling and oversight practices can travel across markets, and the CDSM copyright framework offers a companion baseline for the licensing conversation.

Reflections How can European newsrooms help partner organisations elsewhere adopt AI Act–compatible transparency standards, without imposing them? Which elements of the AI Act should the industry itself codify into operational rules before regulators need to enforce them?

25.11 — Musicians, artists, and the limits of licensing without consent

Bloomberg reports that musicians are pushing back against studios using their songs to train AI, noting that record labels can license their catalogues but that AI companies cannot manipulate specific tracks without artist consent. The distinction between rights held by labels and rights that remain with performers is central to the current dispute, and it mirrors the position taken by news agencies on their own material.

MIT News describes CSAIL research identifying “attribution decay”, a phenomenon in which large training datasets make individual images increasingly difficult to trace inside a generative model’s output. Zheng Dai and Professor David Gifford show that at sufficient scale, removing a single image, an artist’s entire body of work, or every photograph of a person can leave outputs unchanged; Gifford argues that the findings raise questions about copyrightability, author compensation, and whether models should be revised to produce outputs guaranteed to be unattributable to individual creators.

Read together, the two pieces sketch a legal frontier. Consent and licensing remain the working framework in Europe, and the CDSM’s opt-out mechanism gives a first structure, while technical work on attribution and provenance is likely to shape how far claims can be brought and priced. This is law in search of equilibrium, in the strict sense: claims will establish the boundaries, and negotiations will set the workable prices.

Reflections How should European copyright practice adapt when attribution to individual works becomes technically difficult at scale? What licensing structures can protect performers and journalists whose consent has not been secured by the entities that hold their catalogue or archive?

25.12 — Licensing shows up in citation shares, and Google offers publishers a new button

A Press Ranger and OtterlyAI study of 129.3 million AI citations across seven platforms finds that publishers with OpenAI licensing deals receive 48% more citations per page on ChatGPT than unlicensed publishers, and 46% more on average across all platforms. News accounted for 7.2% of AI citations, with Future plc, Forbes, People Inc., Condé Nast, and Hearst capturing 69% of citations to licensed publishers. OpenAI is the only licensor showing a clear home-platform advantage; Google-licensed publishers see slightly lower citation rates on Google AI Overviews, and Perplexity-licensed publishers reach parity on Perplexity.

TechCrunch reports that Google is giving publishers a “Preferred Sources” button they can embed on their sites, allowing readers to mark them as favourites across Search, Discover, and Google News. Google cites earlier studies showing readers are twice as likely to click through when a preferred source is available; more than 345,000 unique sources had already been selected by users as of May.

For strategy leaders, the numbers matter more than the labels. Citation share and referral traffic are becoming the currency of the AI-mediated web, and licensing agreements now have measurable effects on both. The Columbia Journalism Review’s account of the Axios–OpenAI three-year deal, funding thirteen local newsletters and training OpenAI’s models on Axios’s free content, and the NewsGuild’s coordinated “No Slop in Our Shops” protest at USA Today Co. against a Palantir partnership show the political economy of these arrangements from two sides. Deals are becoming a business tool; they are also becoming a governance and labour question.

The Editor and Publisher piece on the agentic-interface era, and the argument that the filter bubble is morphing into something new as agents mediate access to information, completes the picture. As agents choose which sources to consult, the value of being a preferred, licensed, and verifiable source will compound.

Reflections If licensing deals now show a measurable citation premium, how should mid-sized publishers evaluate the trade-off between short-term revenue and long-term dependency? Which governance mechanisms should be built into publisher–platform deals to protect editorial independence and staff conditions?

VI. On press freedom and the fragility of public information

Two items remind us that the political economy of information rests on the safety of journalists and the credibility of investigation.

25.13 — Bangladesh criminalises press-conference questions

The Committee to Protect Journalists reports that Bangladesh’s International Crimes Tribunal has issued fresh warrants for journalists Shyamal Dutta, Mozammel Babu, and Farzana Rupa, alleging that questions they asked at a July 2024 press conference incited a crackdown on demonstrators. All three have been detained since 2024, and, if convicted, could face the death penalty because the Tribunal has no sentencing guidelines for specific offences. CPJ Asia-Pacific Program Coordinator Kunal Majumder called the case “outrageous and deeply troubling” and said editorial decisions “cannot be criminalized”.

This is a European story too. Safety of journalists is one of the three challenges named in the EANA anniversary statement, and international solidarity through alliances such as EANA and WAN-IFRA is one of the few instruments that can raise the political cost of such prosecutions.

For readers of this dispatch, the case is a reminder that AI governance and press freedom belong on the same agenda. A weakened press in one region reduces the supply of verified information for everyone.

Reflections Which international mechanisms are most effective when the act of asking a question is treated as a criminal offence? How should European agencies and publishers structure their solidarity work when press freedom risks converge with AI governance risks?

25.14 — Investigative journalism and the ground it stands on

Craig Unger’s Substack essay argues that investigative journalism has lost much of its impact because Americans no longer share a common factual reality, while social-media algorithms, silos, and weakened institutional checks prevent revelations from producing accountability. He contrasts the effects of Seymour Hersh’s My Lai reporting, the New York Times’ publication of the Pentagon Papers, and Woodward and Bernstein’s Watergate coverage with the political response to more recent scandals, and points to media consolidation, billionaire ownership, and platform capture as structural changes.

Whatever one makes of the argument as political analysis, its editorial implication is important. When information ecosystems fragment, the marginal value of verified, professionally produced journalism increases, and so does the responsibility of institutions that supply it. The Audiencers’ article on synthetic audiences, and its rule to use them “to narrow, never to decide”, makes the same point at the level of research method: AI can help newsrooms sharpen questions, and human evidence must confirm the answers.

For CEOs and editors, this argues for renewed investment in investigation as an infrastructure, not a project, and for a public case that ties AI, licensing, and press freedom into one coherent story.

Reflections If revelations no longer move the political needle, what new formats and partnerships can rebuild the link between investigation and accountability? How should synthetic-audience tools be integrated into editorial planning without displacing the human research that anchors quality journalism?

Conclusion

The Strasbourg text of 1956 was written by agencies that had lived through the collapse and rebuilding of European information. Their successors marked the anniversary this week by naming the tensions of 2026: copyright and provenance, AI-driven disinformation, and the safety of journalists. The choice of these three, rather than a longer list, is itself a statement of priorities.

Around that anchor, the week’s material describes an ecosystem moving from anecdote to measurement. Pew tells us how much of the web is now written with AI. Press Ranger and OtterlyAI tell us how much a licensing deal is worth in citation share. The EU AI Act tells us how transparency will be enforced from December, and the CDSM tells us how consent and remuneration should be framed in copyright. Backstory and SynthID tell us how provenance will be checked. Each of these is a piece of infrastructure that publishers and agencies will need to master.

Two things follow. First, the “no consent, no content” principle set out by EANA is more than a slogan; it is the European answer to a global question about training data, and it belongs at the heart of the licensing conversation. Second, the human premium documented by employers and researchers is not only an HR argument; it is a strategic position for the quality newsroom and for press agencies as critical information infrastructure.

The founding text of 1956 was signed in a specific room in Strasbourg. Seventy years on, the practical work of that same cooperation runs through workflow decisions, licensing negotiations, transparency reports, and training programmes. The anniversary matters because it names the direction; the next decade will be measured in how consistently the direction is held.

For CEOs, policymakers, and newsroom leaders reading this dispatch, the immediate question is straightforward. Which of the mechanisms surfaced this week, transparency, licensing, verification, skills, and cooperation, do our organisations already run well, and which need to be built before December, before the next election cycle, and before the next anniversary?

The Terms of Access

Cyber Territories — Dispatch from the Frontiers

(Synthetic Image)

What does access to journalism mean when the first reader is increasingly a machine and the first answer increasingly appears before a reader reaches the source?

This week’s signals concern a changing relationship between publication, discovery and remuneration. French newspapers have asked the competition authority to examine Google’s AI summaries. Publishers are testing how content can be made legible to agents. AI companies are discussing licences for current news. Australia is refining a bargaining framework that treats journalism as economic infrastructure.

The common issue is control over the conditions of access. Search, voice assistants and agentic systems can make reliable information easier to find. They can also absorb the audience relationship, the context surrounding a report and the economic return that supports its production.

For news organisations, regulators and technology companies, the work is therefore practical. It concerns rights management, attribution, auditability, product design and a sustainable price for scarce professional reporting.

I. The Answer Surface

24.1 — French newspapers challenge Google’s AI summaries

Source

• French media asks French antitrust watchdog to act against Google’s AI summaries — Reuters
  https://www.reuters.com/world/french-media-asks-french-anti-trust-watchdog-act-googles-ai-2026-08-11/

• Google AI summaries: French dailies file complaint with competition authority — Euronews
  https://www.euronews.com/next/2026/08/11/google-ai-summaries-french-dailies-file-complaint-with-competition-authority

• French newspapers target Google over AI summaries — France 24
  https://www.france24.com/en/france/20260811-french-newspapers-target-google-over-ai-summaries

Dispatch

The Alliance de la Presse d’Information Générale, representing nearly 300 French newspapers, has lodged a complaint with France’s competition authority concerning Google’s AI-generated search summaries. The association argues that Google introduced AI Overviews and AI Mode in France without prior authorisation or dedicated remuneration under the French framework for press publishers’ rights. The complaint also raises concern about the effect of answer-led search on visits to publishers’ sites.

The case puts a concrete legal question before a European regulator. Press publishers’ rights were designed for digital uses of protected press publications, while generative summaries reshape the practical meaning of use, display and referral. The relevant assessment includes the way the product is presented, the degree to which it substitutes for the source, the availability of meaningful attribution and the commercial consequences of a publisher’s choices.

Google has the scale and responsibility of a principal gateway to public information. A constructive equilibrium requires that the largest gateways can improve discovery while publishers retain a workable ability to control and monetise the professional material on which those services rely. Legal claims and commercial agreements can together define that equilibrium.

Reflections

• What evidence should establish when an AI summary substitutes for a visit to the underlying publication?

• How should a competition authority assess the practical value of an opt-out option for a publisher?

• Which elements of attribution make a source link meaningful for readers and commercially useful for publishers?

24.2 — The web develops a response to answer-led discovery

Source

• Google’s AI Is Killing the Web. Now the Web Is Fighting Back — Bloomberg Opinion
  https://www.bloomberg.com/opinion/articles/2026-08-10/google-s-ai-is-killing-the-web-now-the-web-is-fighting-back

• How AI poses a threat to journalism already weakened by 20 years of digital upheaval — Le Monde
  https://www.lemonde.fr/en/economy/article/2026/08/11/how-ai-poses-a-threat-to-journalism-already-weakened-by-20-years-of-digital-upheaval_6756369_19.html

Dispatch

Parmy Olson argues in Bloomberg that AI systems now play a far larger role in web activity and that Google’s position in search gives it unusual influence over the transition from links to answers. Le Monde places this change in the longer history of digital disruption to journalism, where advertising, audience data and discovery have already moved towards a small number of major intermediaries. The present phase concerns the use of reporting within a response that may satisfy the user before a publisher’s page is opened.

This changes the scope of publisher strategy. Website traffic remains important, yet it captures only one form of value. A professional publisher also needs to know whether its reporting has been retrieved, quoted, grounded, cited, displayed in a voice answer or used within an agentic task, and under which contractual terms each use takes place.

The web’s original referral structure supported both discovery and accountability because a reader could follow a claim back to its publication. AI-mediated access requires equivalent mechanisms in a different interface. The commercial issue and the democratic issue are closely connected: public information remains stronger when readers can trace a claim to a responsible source.

Reflections

• Which machine uses of reporting should lead to a referral, a licence, an audit record or a combination of all three?

• How can publishers measure value when a report is consumed within an answer rather than on their own service?

• What responsibilities should major gateways accept for preserving a usable route from answer to source?

24.3 — The backlash against AI content begins to influence product choices

Source

• The AI Slop Backlash Is Actually Having an Impact — WIRED
  https://www.wired.com/story/the-ai-slop-backlash-is-actually-having-an-impact/

Dispatch

WIRED reports that several platforms have begun responding to user frustration with low-quality AI-generated material. The examples include tools to flag or label synthetic output, changes to recommendation systems and restrictions on some forms of AI-generated video or impersonation. The emerging pattern is a market signal from users who value relevance, authenticity and a clear sense of responsibility for what they encounter.

For professional journalism, the useful lesson concerns legibility. A newsroom should be able to show what makes its work distinctive: named editorial responsibility, identifiable sourcing, corrections, editorial judgement and a record that remains available when a claim is challenged. These qualities are operational features of a news product, rather than a nostalgic preference for older production methods.

The public reaction also places a limit on the assumption that volume creates value. AI can lower the cost of producing material, while trust continues to depend on judgement, evidence and accountability. In an information market filled with abundant output, reliable reporting becomes a scarce service.

Reflections

• What information should a publisher disclose about AI assistance while preserving a clear and useful reader experience?

• How can editorial provenance be made visible without becoming a superficial marketing label?

• Which newsroom processes deserve stronger investment because they create trust that automated output cannot readily replicate?

II. The Machine Reader

24.4 — TIME tests advertising for agents and Perplexity blocks it

Source

• Perplexity blocks Time’s ads served to AI agents, calling them “deceptive” — Digiday
  https://digiday.com/media/perplexity-blocks-times-ads-served-to-ai-agents-calling-them-deceptive/

Dispatch

Digiday reports that Perplexity blocked advertising from TIME’s markdown pages from influencing its AI system. The material was designed for consumption by AI agents, and Perplexity described the practice as deceptive. The episode concerns an early attempt to monetise a machine-readable layer of publishing, at the point where agentic systems increasingly retrieve content in formats distinct from the page designed for human readers.

The commercial instinct is understandable. When agents become an audience interface, publishers will seek advertising and sponsorship formats that travel with their content. Yet an advertising disclosure that works in a visual page design may lose its meaning when extracted, reformatted and incorporated into an AI-generated response.

This creates a new integrity question for the agentic web. Commercial messages need to remain identifiable as commercial messages across every relevant representation, including markdown, APIs, retrieval systems and generated answers. Advertising standards, editorial standards and platform policy will increasingly meet in the same technical layer.

Reflections

• What machine-readable disclosures are necessary for sponsored material to remain recognisable after retrieval and synthesis?

• Which party should carry primary responsibility when advertising is transformed by an AI intermediary?

• How can publishers develop agent-facing revenue formats that protect audience trust?

24.5 — Apple considers paying publishers for current information in Siri

Source

• Apple in Talks to Pay Publishers to Improve AI-Powered Siri — The Wall Street Journal
  https://www.wsj.com/business/media/apple-in-talks-to-pay-publishers-to-improve-ai-powered-siri-0641f64b

• Apple in talks to pay publishers to provide Siri with current news — TechCrunch
  https://techcrunch.com/2026/08/13/apple-in-talks-to-pay-publishers-to-provide-siri-with-current-news-report/

Dispatch

The Wall Street Journal reports that Apple is in discussions with publishers over arrangements to use current news and information in an upgraded AI-powered Siri. TechCrunch reports that the company has considered a substantial budget for such arrangements. The discussions indicate that current reporting, with identifiable provenance and continuing editorial updates, is being treated as a product input with a distinct commercial value.

A usage-related licence can align remuneration with the practical contribution of reporting to a service. It also creates a demanding governance requirement: the party operating the model has the clearest view of retrieval, ranking, citation and user interaction. Publishers therefore need reliable reporting, clear definitions of use and an independent basis for reviewing the calculation of payment.

The important distinction lies between historic material used for model development and fresh material used to inform people about current affairs. The latter has a short period of high value and depends on professional reporting infrastructure. That combination creates a strong basis for specialised licensing models.

Reflections

• Which events should count as a remunerable use of a publisher’s material in an AI assistant?

• How can a publisher verify usage data when the AI provider controls the technical environment?

• What rights should remain with the publisher when current reporting is licensed for an answer service?

24.6 — News agencies can become the provenance layer for agents

Source

• Japanese publishers are fighting imposter news sites with a cryptographic signature — Nieman Lab
  https://www.niemanlab.org/2026/08/japanese-publishers-are-fighting-imposter-news-sites-with-a-cryptographic-signature/

• AI newsrooms are breaking news now. Haha, I’m in danger — WIRED
  https://www.wired.com/story/ai-newsrooms-are-breaking-news-now-haha-im-in-danger/

Dispatch

Nieman Lab describes how Japanese publishers are using cryptographic signatures to help distinguish their reporting from imitation sites. WIRED examines the prospect of AI-produced news operations and the pressure this creates for established newsrooms. Together, the two pieces concern the same operational reality: large volumes of plausible information can now be produced or imitated with limited cost, while readers and systems need reliable ways to identify accountable sources.

For agencies and publishers, structured reporting offers an important advantage. Time stamps, bylines, correction records, rights information, source metadata and editorial provenance can become machine-readable signals. These are valuable to human users and equally valuable to systems that must decide what to retrieve and cite.

A cryptographic signature does not create trust by itself. It can support a broader verification architecture in which technical provenance, editorial responsibility and legal accountability reinforce each other. The resulting product is richer than an article feed: it is a trustworthy source layer for the information systems that will answer many users’ first questions.

Reflections

• Which provenance data should news agencies make available to AI systems as a standard service?

• How can publishers ensure that technical authentication remains understandable and useful to ordinary readers?

• What common standards would allow professional news organisations to distinguish their reporting from synthetic imitations at scale?

III. The Price of Presence

24.7 — Australia refines the bargain between platforms and news

Source

• Australia to make tech companies pay more outlets for content under reworked media law — Reuters
  https://www.reuters.com/business/australia-make-tech-companies-pay-more-outlets-content-under-reworked-media-law-2026-08-12/

Dispatch

Reuters reports that Australia has revised its proposed media framework to require large technology companies to strike agreements with more local news outlets in order to reduce a levy on Australian digital advertising revenue. The framework uses a 2.5% levy and sets conditions for qualifying agreements, including a broader range of participating media organisations. Reuters also reports that a portion of the funds would support Australian Associated Press.

The design is significant because it combines a statutory incentive with commercial negotiation. The law defines the conditions under which major platforms contribute to the news ecosystem. Agreements then determine the operational terms, including the publishers involved, the rights used and the value assigned to those uses.

The explicit role for AAP recognises the infrastructure function of a news agency. Agencies gather, verify and distribute reporting that supports many outlets, including smaller titles that cannot build every reporting capability alone. Media policy that values diversity of supply needs to account for this shared professional infrastructure.

Reflections

• How should a remuneration framework recognise the role of agencies in supporting a diverse news ecosystem?

• Which incentives encourage broad participation without treating all forms of content as economically equivalent?

• What lessons can European policymakers draw from a model that combines legal obligations with negotiated agreements?

24.8 — Publishers seek growth beyond the website

Source

• Non-website audience is focus of revenue growth for new Business Insider CEO — Press Gazette
  https://pressgazette.co.uk/publishers/digital-journalism/non-website-audience-is-focus-of-revenue-growth-for-new-business-insider-ceo/

• The Post launches Hamilton, a new way to find and understand the news — New York Post
  https://nypost.com/2026/08/11/media/the-post-launches-hamilton-a-new-way-to-find-and-understand-the-news/

• New York Post launches Hamilton AI chatbot — Axios
  https://www.axios.com/2026/08/11/new-york-post-hamilton-ai-chatbot

Dispatch

Press Gazette reports that Business Insider’s new leadership sees audiences beyond the website as a central area for revenue growth. In the United States, the New York Post has launched Hamilton, an AI chatbot designed to help users find and understand news from the publication. Axios describes the tool as part of a broader effort to place a publisher’s journalism within an interface designed around direct questions and responses.

These initiatives reflect a calculated choice to meet audiences on the services and formats they use. News organisations can develop products in apps, newsletters, video, audio, messaging, voice and AI interfaces while retaining their own editorial identity and commercial relationship. The critical point is the design of that relationship: attribution, access to user insight, subscription pathways, advertising integrity and rights control all require deliberate choices.

An owned AI interface offers a publisher a practical laboratory for these choices. It can demonstrate how a news organisation wants its journalism to be retrieved, explained, cited and corrected. This allows publishers to negotiate with external AI services from experience rather than from abstraction.

Reflections

• Which audience surfaces create a durable relationship with a publisher rather than only temporary reach?

• How can an owned AI product strengthen subscriptions, trust and editorial differentiation at the same time?

• What data and governance safeguards should apply when a publisher uses reader interaction to improve an AI news service?

24.9 — Bot access becomes a commercial and cybersecurity issue

Source

• European publishers are getting hit harder by AI bot scraping, report finds — Digiday
  https://digiday.com/media/european-publishers-are-getting-hit-harder-by-ai-bot-scraping-report-finds/

• Stopping the bad bots blitz — WAN-IFRA
  https://wan-ifra.org/2026/08/stopping-the-bad-bots-blitz/

• 3 separate audits, 3 research teams, 1 common failure — WAN-IFRA
  https://wan-ifra.org/2026/08/3-separate-audits-3-research-teams-1-common-failure/

Dispatch

Digiday reports that European publishers face particularly intense AI-bot scraping, adding technical costs and pressure on publishers’ digital infrastructure. WAN-IFRA’s work on harmful bots and recurring weaknesses in AI systems adds a related concern: automated systems can create operational risks when access controls, authentication and content-use policies do not keep pace with the volume and sophistication of automated requests.

Bot management has therefore moved from a technical housekeeping task to an executive concern. It affects infrastructure expenditure, licensing enforcement, cybersecurity, measurement and the integrity of a publisher’s commercial model. A bot that retrieves at scale may be a legitimate licensed customer, a search crawler, an unauthorised scraper or an adversarial actor, and each category requires a different response.

The useful objective is granular control rather than a universal block. Professional publishers need to know who accesses their material, for what purpose, at what volume, under which identity and with which contractual permissions. Access control becomes part of both cybersecurity governance and rights governance.

Reflections

• Which categories of automated access should a publisher identify and govern separately?

• How can media organisations combine bot controls with commercially viable access for trusted AI partners?

• What board-level indicators would show whether bot activity is creating material financial, legal or security exposure?

Conclusion

The central issue across this week’s developments is the quality of access. AI can make information more immediate and more useful, especially when it helps people understand a complex event or locate reliable reporting. The same systems determine how much of the source remains visible, attributable and economically viable.

The illustration accompanying this Dispatch captures the emerging division in a useful way. One entrance presents the open web as a human environment, supported by browsing, reading and connecting. The other presents machine access as authenticated, metered and governed through an API. The image should be read as an operational warning: the two forms of access are developing different rules, and public-interest journalism needs a coherent place in both.

For publishers, the immediate agenda is practical. Define machine-readable rights, establish provenance standards, segment bot access, build direct AI products where useful and insist on auditability in licensing agreements. For agencies, structured and verified reporting can become a core input for trustworthy retrieval and citation systems. For regulators, the task is to ensure that rights, competition rules and transparency requirements remain effective when the interface shifts from a link list to an answer.

The aim is a healthy equilibrium. Quality journalism is scarce, costly and accountable. Its use in AI systems can create real value for technology companies and users, provided the terms preserve attribution, editorial integrity, source access and proper remuneration. Litigation can establish legal boundaries, licensing can establish workable prices and innovation can create products that serve readers across the human and machine layers of the web.

The Machine as Reader

Cyber Territories — Dispatch from the Frontiers

The machine as reader. (Synthetic Image)

For most of the history of journalism, the reader was the point. This week’s signals describe a system in which the first reader is a machine, and in which the physical record itself is being converted into material for that machine to read.

Two events set the frame. Google withdrew a feature that let anyone generate plausible satellite imagery of real places, which is the record of the physical world. Anthropic has been buying and destroying printed books after scanning them, which is the record of human thought. Both turn something durable into something fluid, while the institutions that verify durable things are asked to keep up.

Around them, the commercial layer is settling. Australia has finalised a charge on platform advertising revenue, Getty has licensed display rights to OpenAI, and UK publishers are signing confidential agreements with Google while criticising the terms in public. Underneath sits the audience: Australians now reach news through social platforms more often than directly, The New York Times reports exposure to declining search traffic, and the largest American newspaper group has hired Palantir to turn anonymous visits into known relationships.

The theme is intertwining. Physical and digital record, human and synthetic reader, editorial and commercial decision are now bound together. The professional task is deciding which of those bindings a news organisation accepts.

I. On the Record of the World

Three signals about what remains verifiable when the underlying record becomes generative.

23.1 — Google withdraws synthetic imagery from Google Earth

Google removed an image-generation feature from Google Earth about a day after launch, saying people were “sharing screenshots of generated imagery that appear to violate our policies” and that it would roll the feature back “while we work on implementing stronger guardrails”, as The Independent reported. Open-source researcher Henk van Ess told NPR he had generated refugees at the Mexican border, a nuclear plant in Iran and a hospital with a bomb crater in Gaza, and that “nothing was refused”.

Satellite imagery occupies a specific place in verification work, treated by investigative desks and press agencies as close to independent evidence precisely because it is expensive to produce and hard to fake at scale. A consumer feature that generates convincing variations changes the cost of forgery for an entire category of reporting, and watermarking protects the file rather than the screenshot that circulates.

The speed of the withdrawal is the useful part. Platform product cycles now move faster than the verification community can assess them, and the correction depended on public criticism arriving within hours. A guardrail applied after global release amounts to testing on the information ecosystem.

Reflections When the imagery used to confirm events becomes generative by default, what evidence does a newsroom accept as sufficient? Which verification capabilities should press agencies build in-house rather than source from the companies shipping these features?

23.2 — India presses platforms and rebuilds fact-checking around AI

India’s technology ministry has asked Meta for a compliance roadmap on moderation, data localisation and faster removal of flagged material, including algorithmic changes to slow manipulated content and deepfakes, Mathrubhumi reported. In parallel, the Press Information Bureau’s Fact-Check Unit lost its proposed statutory status after the Bombay High Court struck down the relevant IT Rules provision, and the government is now building an AI chatbot for real-time claim verification, an archive of past checks and a detection system for manipulated content, according to India Today.

The combination is instructive. A court removed the legal basis for a state fact-checking authority, and the executive responded with a technical capability that requires no statute. The unit published more than 2,900 fact-checks by April, and accreditation rules already allow debarment of journalists for two to five years, so the capacity being automated is substantial.

For European readers the comparison is procedural. The AI Act and the DSA place obligations on providers through defined processes with review; an administrative verification capability built into government systems concentrates the same function without them. The question is which institution may declare something false, and how that declaration can be contested.

Reflections Where should the authority to label a claim false sit when courts limit statutory fact-checking bodies? How does a news organisation cooperate with state verification systems while preserving independent editorial judgement?

23.3 — Anthropic converts printed books into training material

Anthropic has been buying used and rare books in bulk, cutting off their spines, scanning the loose pages and sending the remains for recycling, a practice examined critically in The Spectator. The article links the destruction to last year’s $1.5 billion copyright settlement, in which a United States court accepted training on text while finding that Anthropic had acquired titles without authorisation, and notes that destroying the physical copy was treated as a way to avoid retaining a duplicate.

That reasoning belongs to a United States framework and should be read carefully in Europe. Under droit d’auteur and the CDSM Directive the questions are lawfulness of the source, applicability of text and data mining exceptions and effectiveness of reservation of rights, and none of them turn on whether the physical object survives the scan. Lawful acquisition of a copy and lawful use of the expression it contains remain separate matters.

The cultural point is the stronger one. A library holds a book so a later reader can return to it; a training pipeline holds statistical traces so a model can approximate it. Only one of those allows a future disagreement about what the text actually says.

Reflections What obligations should a company accept toward the physical cultural record it consumes for training? If out-of-print works survive only as model weights, who is responsible for the loss?

II. On the Price of Access

Three signals about law and licensing working as two halves of the same process.

23.4 — Australia finalises its charge on platform advertising revenue

Labor has finalised the News Bargaining Incentive and narrowed its base to Australian digital advertising revenue, reported by the Australian Financial Review, which also noted that the carve-out for professional networking sites has been reversed, bringing LinkedIn within scope. Reuters set out the mechanics: 2.5%, up from 2.25%, on platforms with more than A$250 million in local revenue, offset by qualifying agreements with at least six publishers rather than four, with larger offsets for regional outlets and all proceeds directed to the news sector. Standalone AI services remain outside the scheme.

This is the clearest recent example of law and negotiation operating together. The charge establishes the boundary, the offsets establish the price, and the design makes a commercial agreement cheaper than the statutory alternative, with Treasury expecting roughly A$250 million a year to move toward journalism.

Two features deserve attention in Brussels. Including LinkedIn treats professional networks carrying news as distribution; excluding standalone AI services treats generative systems as a separate regulatory problem. European legislators considering remuneration mechanisms face both decisions in a market that already has a neighbouring right.

Reflections If a statutory charge sets the effective floor for licensing, what remains for publishers to negotiate? Should generative AI services be brought inside a bargaining mechanism, or governed under a separate instrument?

23.5 — UK publishers weigh confidential Google terms while search changes shape

Press Gazette describes a prisoner’s dilemma among UK publishers, with major titles signing confidential two-year agreements covering AI training, grounding and AI Mode on take-it-or-leave-it terms including non-disclosure and no-sue clauses, earning single-figure millions per year. One source called them “no-sue deals” that amount to “renting peace”; Jason Kint of Digital Content Next observed that Google seeks what it needs “without putting a direct monetary value to the content licensing”, while Madhav Chinnappa argued for a collective access model built on APIs and subscriptions.

The product side moved in the same week. Google has begun placing a Top Stories carousel inside AI Overviews for news queries, appearing in roughly 15% to 17% of trending news searches in the US and UK according to Newzdash analysis reported by Press Gazette, with publishers using the Search Console generative-AI exclusion likely losing eligibility for that placement. Newzdash founder John Shehata called it “a double-edged sword for publishers”.

Together these describe one mechanism. The CMA ruling gave publishers a right to opt out of generative uses without harming conventional rankings; product design attaches a traffic cost to exercising it and commercial terms attach a cash cost to litigating it. Rights that are expensive to exercise are exercised rarely, which is the test any remedy has to pass.

Reflections What conditions would make a collective licensing model more attractive than an individual confidential agreement? How should a regulator measure whether an opt-out right is genuinely available in commercial practice?

23.6 — Getty licenses display rights and finds a new revenue line

Getty Images and OpenAI announced a multi-year, non-exclusive display partnership in late June, allowing OpenAI to show licensed Getty pictures to ChatGPT users with attribution while granting no model-training rights, according to a MarketBeat analysis carried by The Globe and Mail. The structure mirrors Getty’s October 2025 agreement with Perplexity. The context matters: around $2.0 billion of total debt, roughly $194 million of expected cash interest for 2026, shares at $0.58 in June against an active NYSE compliance notice, and a pending $3.7 billion merger with Shutterstock conditioned by the CMA on divesting the UK and global editorial business.

The separation of display rights from training rights is the transferable idea. It creates recurring, auditable revenue without transferring the asset that generates future value, and it prices distribution rather than ingestion. For press agencies, whose archives combine editorial value with indemnified provenance, that distinction applies directly.

The financial detail carries a caution. Getty negotiated under balance sheet pressure, and terms agreed under pressure set reference points for everyone who negotiates later. Scarcity supports price only when the seller can afford to wait.

Reflections Which rights in an archive should be licensed for display, and which should remain unavailable at any price? How do rights holders avoid setting sector-wide reference prices during periods of individual financial strain?

III. On the Audience That Moves

Three signals about where readers arrive from, and what publishers are building in response.

23.7 — Social access overtakes direct visits in Australia

The twelfth Digital News Report: Australia found social media has passed direct visits to news sites and apps for the first time, with television at 57%, social at 56% and direct access at 52%, reported by AdNews. Among 18 to 24 year olds, 60% have never read a printed newspaper and TikTok use for news reached 48%, up twelve points. Nine percent now use chatbots for news, while 49% distrust news from chatbots and 51% distrust news from social media.

The payment data complicates the decline narrative. Australia records the highest share of consumers paying for digital-only news brands at 35%, and trust in self-selected sources rose five points to 54% against 21% for social and 19% for chatbots. Readers distrust the intermediary while using it, and pay for brands they have chosen deliberately.

That defines the commercial opportunity. Distribution has moved to surfaces where trust is low, and value accrues to organisations that convert an untrusted arrival into a chosen relationship. Creators now serve as a news source for 43% of Australians, which is the competitive set for attention, though the standards set is a different one.

Reflections What does it take to convert an arrival from a low-trust surface into a deliberate subscription decision? Which audience segments justify continued investment in platform distribution, and which do not?

23.8 — The Times leans into video while a large chain turns to data infrastructure

The New York Times added 280,000 digital-only subscribers in the second quarter, down from 310,000, reaching 13.3 million subscribers and $538 million in subscription revenue, with CEO Meredith Kopit Levien saying the company “isn’t immune” to an ecosystem shaped by a small number of large technology companies, according to Nieman Lab. The response is video: eight video journalists hired in January, twelve more roles open, a new Shows tab, and a 10% rise in adjusted operating costs. The Times also disclosed $32.9 million in generative-AI litigation costs since early 2024.

USA Today Co., owner of more than 200 local titles, has partnered with Palantir to analyse and monetise audience behaviour, with CEO Mike Reed describing the goal as converting anonymous interactions into “known, orchestrated first-party relationships”, also reported by Nieman Lab. Unique visitors fell from 180 million to 158 million in a quarter, attributed to lower search referrals rather than lower demand, and Reed said he would prefer a fair licensing deal with Google but would block access if necessary.

The two responses illustrate the choice available to publishers of scale. One invests in a format intermediaries cannot easily summarise; the other invests in knowing its audience more precisely, accepting a supplier whose government work makes this a governance decision at board level rather than a procurement decision at management level.

Reflections Which audience data capabilities should a news organisation own outright rather than source from a vendor? How much of a shift toward video is defensible on editorial grounds when the driver is algorithmic visibility?

23.9 — Building on owned ground while the optimisation industry reorganises

Puneet Gupt, COO of Times Internet, told WAN-IFRA’s Digital Media India event that publishers spent years treating the crawler as their most important reader and “lost track of who our loyal readers were”, summarising his message as “don’t build on borrowed land”. He called this “an ecosystem change” rather than an algorithm change, said “the runway is short”, and set out three requirements: AI-native operations designed as a complete redesign, direct reader relationships built on consent-based data, and trust-led revenue, since “trust is a conversion multiplier in whatever product you create”.

The optimisation industry has already reorganised around the new surfaces. The Verge examined how the SEO sector is attempting to influence AI answers through generative engine optimisation, self-serving listicles and, in cases documented by Microsoft, hidden instructions behind summarise buttons. Google updated its search spam policies in May 2026 to cover manipulation of generative responses, which confirms the practice is widespread enough to require enforcement.

The economics are clear. When visibility inside an answer becomes the scarce good, an industry forms to purchase it, and the cost of maintaining a trustworthy answer layer rises for whoever operates it. Publishers holding a direct relationship with readers hold the one asset this market cannot manufacture.

Reflections Which parts of a publisher’s business currently depend on ground it does not own? How should a newsroom respond when competitors invest in influencing AI answers about their own coverage?

IV. On the Craft That Holds

Three signals about newsroom practice where AI has become routine.

23.10 — Reuters sets speed against editorial responsibility

An analysis of Reuters’ newsroom AI describes assisted reporting in which technology surfaces material while “a journalist must establish what it means”, framing the conflict as automated speed against accountable editorial judgement. The pillars remain experimentation, journalist approval on the principle that “a Reuters story remains a Reuters story”, disclosure of material AI use, and verification, with generative systems prohibited from creating or enhancing news imagery. FactGenie reached 150 journalists and halved the average time to send non-corporate alerts, though error and correction rates remain unpublished.

The industry data gives context. A 2026 Reuters Institute survey found 64% of media leaders consider back-end automation important while 13% describe current initiatives as transformational, and 67% report that AI has reduced no roles. A study of 1,004 UK journalists found 60% reporting some integration, 44% with rules for human oversight and 27% with guidelines on bias and fairness.

The governance insight belongs in every management team. Human review functions as a safeguard only when the reviewer has time, access to the source, an understanding of how the system fails and the authority to reject its output. Sequential automation moves an early error through every later stage before anyone sees it, which is the specific risk agentic workflows introduce at scale.

Reflections What would a meaningful disclosure standard look like when AI use is described as material? Which stages of a reporting chain must always show their sources before a human approval carries weight?

23.11 — African newsrooms and a Korean regulator build AI around the desk

Briefly News, part of Legit, described the development of Nova at INMA’s Africa Newsroom & Tech Summit, where 25 to 40 writers producing up to six stories a day are supported by one or two copy editors. An early custom GPT produced hallucinated corrections, so writers spent “more time looking for the error that wasn’t there than actually fixing real errors”, leading to a rebuilt Chrome-based Moffield Copilot beside the CMS with media-law and policy compliance checks. Editors doubled the stories they could review, and readers remained the constraint: “they can tell when it is AI-created”.

Punch Newspapers set a stricter boundary at the same summit, reported by Công Luận: AI may support ideation, research, summarisation, transcription and proofreading and may generate SEO metadata after publication, while writing articles and editing news photographs remain prohibited. Editor Oyetunji Abioye put it plainly: AI assists the journalist, and every published article has a journalist who answers for it.

Korea’s Internet Newspaper Ethics Committee has taken the same logic into self-regulation, deploying an AI keyword-monitoring system in August that collects articles from 850 participating internet newspapers every two hours and flags profanity, discriminatory expressions, unclear sourcing and crime or suicide reporting. Review-office head Park Young-rye said the purpose is to support accumulated standards and specialist judgement rather than replace human review; detection triggers a human reading, with decisions taken in weekly and monthly meetings.

Reflections Which editorial tasks should remain off limits to automation regardless of measured efficiency gains? When a self-regulatory body automates detection, how does it prevent the tool’s categories from becoming the standard itself?

23.12 — A new book argues AI is accelerating a system already under strain

Journalism in the Age of AI argues that current developments are “largely accelerating a broken system” that has weakened journalists and public trust for decades, tracing AI across ideation, sourcing, verification, storytelling and distribution in an excerpt published by Nieman Lab. CalMatters uses AI to track state legislators and generate tips that Sisi Wei says “would have taken a data-savvy political reporter weeks, if not months, to find”, while Full Fact, Maldita.es and the European Fact-Checking Standards Network run a multilingual system that extracts claims from short-form video.

The failure cases are equally specific. Die Aktuelle published a fabricated Michael Schumacher interview and settled with his family, CNET corrected machine-generated stories, and in November 2025 reporters at Suncoast Searchlight found undisclosed AI editing had inserted hallucinated quotations into their drafts. A study of 22 fact-checkers found available tools often failed to fit workflows and increased workloads.

The strategic reading is that AI amplifies whatever structure a newsroom already has. Organisations with defined standards and clear accountability gain capacity; organisations without them acquire faster ways to publish unverified material. The differentiator is institutional, which is the asset a quality newsroom holds and a model cannot reproduce.

Reflections Which existing weaknesses in your editorial process would AI adoption make more visible within a year? What consent should journalists have over AI systems that alter their drafts before publication?

V. On the Machine as Reader

Three signals about what happens when machines read, speak and are believed.

23.13 — Readers rate AI writing highly and increasingly listen to synthetic voices

A study led by Villanova University researchers and published by Cambridge University Press found that participants aged 18 to 81 could not reliably distinguish AI-generated short stories from those by published authors, as the BBC reported. The highest ratings went to stories written by AI and described as human. Senior author Dr Deena Weisberg said this “reveals a bias towards narratives written by real people”, that “public assumptions about AI’s capabilities are increasingly out of date”, and that AI writing “tends to be clearer, more direct and easier to process”. Greater AI literacy improved detection; literary expertise did not.

The audio layer is scaling in parallel. The Columbia Journalism Review examined synthetic narration across the Times, the Post, the Journal, The New Yorker and The Atlantic, noting that ElevenLabs has passed $500 million in annual recurring revenue with early investor talks at a $22 billion valuation. Around 20% of New Yorker subscribers listen to narrated stories, and the Journal’s Read to Me feature was used roughly 5 million times in a year. The New Yorker declines to narrate fiction synthetically, treating it as performance.

Both findings point at the same commercial fact. Clarity and convenience convert, and a machine supplies both at marginal cost. What it supplies less well is the register that carries irony, restraint and judgement, which is much of what a reader pays a brand for.

Reflections If audiences rate machine-written text highly when they believe a person wrote it, what does authorship promise them? Where does synthetic narration serve accessibility, and where does it change the work itself?

23.14 — Chatbots outrank elected representatives on trust

A Bloomberg newsletter reported survey findings that people now regard AI chatbots as more trustworthy than politicians. The underlying work comes from the Collective Intelligence Project’s Global Dialogues, whose published snapshot draws on roughly 4,000 participants across four rounds and finds respondents consistently rating AI as capable of better decisions on their behalf than their government representatives, with the share rising each round. Only public research institutions and family doctors are trusted more than chatbots.

Read alongside the Australian data, where 49% distrust news from chatbots, this produces a specific picture. People assign high trust to a system they use for personal reasoning and low trust to the same system as a news source. Trust is being allocated by function, and the function of interpretation is moving toward an interface owned by a few companies.

For democratic institutions this is a measurable signal well before it becomes an electoral one. Representative bodies now compete for credibility with a product that answers immediately, never appears evasive and has no constituency to manage. The comparison is unfair to the institutions and flattering to the product, which is why it holds.

Reflections What should elected institutions learn from an interface that outperforms them on perceived honesty? How do news organisations report on public trust in AI systems without either amplifying or dismissing the finding?

23.15 — DeepMind names recursive self-improvement as the investment thesis

Jasjeet Sekhon, chief strategy officer at Google DeepMind, told an audience at UC Berkeley that recursive self-improvement, meaning AI systems that build better versions of themselves, is the core investment thesis behind the sector’s capital expenditure, as first reported by The Information. He acknowledged that AI revenues “don’t sustain the capital expenditures we’re making so far” and offered the analogy that steam engines were used to build the next steam engine, against annual Google capital expenditure running at roughly $200 billion.

This is unusually direct and deserves to be taken at face value by anyone negotiating with these companies. The economics of the current buildout are justified by a capability that does not yet exist, so near-term revenue matters mainly as evidence that the thesis remains credible to investors. Content licensing sits inside that calculation as a cost to be minimised, which explains a great deal about the terms on offer.

There is no free lunch in this arithmetic. Capital committed today must be recovered from advertising, subscription, enterprise contracts or the value extracted from material others produced. Publishers negotiating access rights in 2026 are negotiating against a balance sheet that requires the cheapest possible input.

Reflections How should a rights holder price access to a counterparty whose business case depends on a future capability? If the recursive self-improvement thesis proves slower than expected, which of today’s licensing terms will look mispriced?

Conclusion

The reader has changed identity. A crawler reads to index, a model reads to train, an answer engine reads to summarise, a synthetic voice reads aloud, and an optimisation industry writes for the machine that reads first. At the end of that chain a person still decides whether to believe what they are told, and every institution in this dispatch is arguing about who gets paid, credited or held responsible along the way.

Two costs are becoming visible together. One is producing verified information, always high and now competing with material generated at almost no cost. The other is maintaining the record itself, whether satellite imagery reliable enough to confirm an event or a printed book a future reader can consult. Both fall on institutions that were already carrying them, while the value created afterwards accrues elsewhere.

The regulatory responses are converging on a workable shape. Australia has priced access through a charge with commercial offsets, the CMA has established an opt-out right whose practical availability is now being tested by product design, and Kenya proposes effects-based jurisdiction with a risk framework and a public register for high-risk systems. Each seeks an equilibrium between claims and contracts, and none will work if the right it creates is more expensive to use than to give away.

For news organisations the conclusion is narrow and demanding. Own the relationship with the reader, own the verification capability, and treat every agreement with an intermediary as a decision about which part of the business sits on ground belonging to someone else. Press agencies carry an additional obligation, since their material becomes the reference point other newsrooms and, increasingly, machines rely upon.

The robot on the cover of this dispatch is reading a newspaper. Someone still had to write it, verify it, print it and stand behind it, and that remains the part nobody has automated. The institutions that keep doing that work hold the last position in the chain where a claim can be checked against something real.

Patrick Lacroix writes Cyber Territories in a personal capacity. AI tools are used for research and drafting; the author retains full editorial responsibility.

Karl Popper called optimism a moral duty. The future stays open, so the people making decisions this summer are the ones who will shape it. That is a good thought to hold on to in a season of confident predictions about artificial intelligence.

The statue on this page suggests the method. Socrates worked with questions, and he aimed them at the assumptions nobody had bothered to examine. Ask what a system learned from, who paid for that material, what the numbers actually measure, and who carries the consequences when the answer turns out to be wrong.

This week’s links are full of people. Reporters filing from Tehran through power cuts. A retired librarian in Manizales learning to check a WhatsApp message. A model in Shanghai who refuses to sell the rights to his own face. Editors somewhere in Europe reading machine output line by line to see whether it is true.

One thread connects them. Machines produce answers at scale, and the credibility of those answers comes from people who checked something first. The research this week puts numbers on that dependence for the first time, and the market is only beginning to work out what it owes.

I. On Trust and the Measurement of Visibility

Three signals on where credibility in AI answers comes from, and on how poorly the industry measures its own visibility.

22.1 — Research confirms that AI answers inherit the trust of the brands they cite

New research from the Association of Online Publishers, reported in Press Gazette, finds that an AI answer is trusted almost exactly as much as the sources it cites. In an Ipsos survey of 1,000 people in the United Kingdom, a fully trusted source produces more than 90 percent trust in the answer, a neutral source around 25 percent, and a distrusted source around 10 percent. Almost half of the users who trust the cited source click at least one link.

The same survey found that 37 percent of respondents did not realise AI tools can invent information or sources, rising to roughly 45 percent among 45 to 54 year olds. That is a duty of care for the people designing these interfaces as much as for publishers.

Credibility travels from the newsroom into the answer. The accounting has yet to follow it.

Reflections If citation is what makes an answer credible, how should that credibility be valued in a licensing conversation?

22.2 — A critical survey questions what generative engine optimisation actually measures

Olivier Martinez has published a critical survey of generative engine optimisation covering 45 studies from November 2023 to July 2026, also available on SSRN. His finding deserves the attention of anyone currently buying GEO services. The widely cited gains from the foundational paper hold only where a source already appears in a fixed context, and they establish neither organic discoverability nor lasting traffic effects.

Topical relevance and position in the context window are the most reproducible levers. Generic heuristics transfer poorly between engines, and rewrites designed to win citations can damage retrieval. Commercial audits show low overlap between sources, wide variation from one run to the next, and persistent gaps in fidelity.

Within the reviewed literature, no technique demonstrates a stable, cross platform effect on organic discoverability. An industry is buying a measurement practice that resembles a fairground more than a discipline, and repeated measurements with proper controls are the cheapest possible insurance.

Reflections Which evidence would make a visibility claim solid enough to justify a budget line?

22.3 — Compensation models begin to pay for citation

Perplexity has announced the launch partners for Comet Plus, a five dollar subscription included with its Pro and Max tiers, with Condé Nast, Fortune, Le Figaro, Le Monde, the Los Angeles Times and the Washington Post taking part. Publishers are paid according to human and machine interactions with their content. Steven Brill told Press Gazette that NewsGuard AI shares revenue with publishers and cites them prominently, drawing only on the 12,000 sources it rates as reliable and using 64,000 catalogued false claims as guardrails.

Brill adds a security argument to the commercial one. Large models repeat provably false claims roughly a third of the time when prompted about them, which he attributes in part to coordinated flooding of the web with fabricated articles. Licensing verified reporting and defending against that flood turn out to be the same project.

Scarcity alone preserves nothing. Verified journalism holds its price where organisations can demonstrate use, keep attribution intact and negotiate from a position of leverage, individually or collectively. Claims in court establish the boundaries and negotiations establish workable prices; a sustainable equilibrium becomes possible only when both are done well.

Reflections What contractual terms would make interaction based compensation genuinely auditable for a news publisher?

II. On the People Who Carry the Technology

Three signals on newsroom policy, on investment in staff, and on where the training is coming from.

22.4 — Three newsrooms choose three different paths, and staff report the cost

Erik Bucy and Milad Jalalian Ebrahimi describe in Nieman Lab how Reuters, the BBC and the Guardian have each built a different AI posture. Reuters uses Lynx Insight and Fact Genie to publish some financial alerts in six to eight seconds. The BBC runs its tools behind mandatory human approval, labels assisted content with a hexagon and the phrase “How we used AI”, and published its own research with the European Broadcasting Union showing AI assistants misrepresenting news content 45 percent of the time. The Guardian, under a trust structure since 1936, allowed limited generative use only from March 2026, with permission from a senior editor.

The authors name the cost precisely: a verification tax, the hours editors spend checking machine output. Adaptavist found the same pattern among 2,500 professionals, with 42 percent reporting they spend more time verifying AI output than they save and 65 percent feeling nostalgic for how work ran before generative tools. In the same survey, 67 percent would like their organisation to use more AI, which describes a workforce asking for better implementation.

Ownership shapes policy. Wire services optimise for speed, public broadcasters add oversight, reader funded titles protect the human byline. Each choice is defensible, and each comes with its own bill.

Reflections How would you measure the verification tax in your own newsroom before setting an automation target?

22.5 — Investment in people remains the decisive choice

The Guardian is creating 55 new permanent jobs across the United Kingdom, the United States and Australia in video, audio and data journalism, with Guardian Australia becoming a global hub for social media hosts. Katharine Viner describes the ambition as combining a global reputation for impactful reporting with journalism that is more visual, digital, global and experimental.

Mitali Mukherjee, one year into her directorship of the Reuters Institute, makes the same point from Oxford. She says that the most important resource in news is individuals, and organisations need to invest in them. She also offers a useful idea for smaller teams, asking in how many forms a single story can be told, from audio capsules to video, and she is refreshingly protective of the fellows who pass through the institute.

Mukherjee adds one observation worth keeping. Access to information works as a barometer of privilege, and audiences under thirty reach news through social and video surfaces while still trusting organisations with a record of covering a subject well. That record is built by people.

Reflections Which capabilities deserve permanent headcount in a newsroom that also automates production?

22.6 — Practical guidance and training arrive from Vienna, Nairobi and Delhi

The Austrian Lab for Artificial Intelligence Trust, an initiative of the Austrian innovation ministry run with APA, has published sector reports for healthcare and news media with concrete instructions for safe AI use. The media report asks for an explicit strategy on permitted uses, disclosure to the audience following the principle of too much rather than too little, careful source handling, and labelling with reference to the AI Act. It advises against treating AI mainly as a way to cut costs, which is the most valuable sentence in the document.

Training is arriving from several directions. The Thomson Reuters Foundation has tendered for an AI newsroom consultant to mentor five independent Kenyan media over three to four months, including the drafting of editorial guidelines. The Google News Initiative has opened 50 seats for an AI Skills Program 2.0 in Delhi, combining a workshop with six weeks of mentoring.

Who funds the training shapes what gets taught. That makes the governance of these programmes as interesting as their curriculum, and it is worth noting that the Tarbell Center funds reporting on AI under an editorial independence policy that keeps donors out of the selection.

Reflections What safeguards belong around training funded by the companies whose tools are being taught?

III. On Reporting When Institutions Are Weak

Three signals on verification under fire, on journalism as a civic service, and on audiences learning to check for themselves.

22.7 — AFP reports a war in which a quarter of flagged content is synthetic

AFP has published an account of its coverage of the Middle East conflict that began at the end of February, with simultaneous military operations, internet blackouts and coordinated disinformation. Around 25 percent of the content flagged to its fact checking teams was generated or manipulated with AI, including game footage presented as bombing raids, fabricated images of captured pilots and a staged funeral scene that became the most widely shared piece of misinformation of the conflict. Its OSINT teams verified more than 200 videos from open sources and distributed them to 200 television networks.

AFP credits a Dubai bureau maintained for more than 35 years and a network of sources built over the same period, which is what allowed reporting to continue from Tehran through explosions and outages. Presence like that cannot be improvised once a crisis starts.

A press agency functions as critical information infrastructure in every practical sense during a conflict of this kind. When a quarter of the suspect material is synthetic, verification capacity becomes the product.

Reflections What verification capacity does your organisation need before the next synthetic media crisis reaches your own coverage area?

22.8 — Venezuelan journalists coordinate relief when official information is unavailable

Tony Frangie Mawad describes in CJR how two earthquakes forty seconds apart on 24 June damaged or destroyed nearly sixty thousand buildings, injured more than eleven thousand people and killed at least two thousand, with tens of thousands still missing. Telecommunications collapsed and official registries were unavailable. The only functioning records were open source platforms built by civilians.

His magazine Ecosistema created a platform called Manos a la Mesa to connect restaurants with displaced people who needed a meal. Reporting, coordination and relief work happened at the same desk, by the same people, in the same week.

Where public institutions weaken, the reporting function becomes the register of last resort. Any newsroom writing a crisis protocol should read this account before the next one.

Reflections When a newsroom becomes a coordination platform during a disaster, which editorial standards continue to apply?

22.9 — Audiences train themselves to challenge misinformation

In Manizales, where 21 percent of residents are over sixty, the Universidad de Caldas and the local outlet Pensé que ‘VOZ’ Sabías run a 48 hour course in digital literacy against misinformation for older adults. Participants learn how fact checkers work, they role play as misinformation creators, and they take apart a WhatsApp message that has circulated since 2023 claiming the city could reach 50 degrees. More than 2,000 people were certified in the first year. One of them, a 69 year old retired librarian, says she learned to question what she reads.

The Public Interest News Foundation puts a related question to a much larger institution, asking whether engagement should become the BBC’s newest public purpose. The proposals include accountability sessions where audiences meet the journalists, citizens’ agendas that let the public set election questions, and partnerships with organisations already trusted in their communities. Trust follows from giving power away and keeping promises.

Two ends of the same system. One teaches the audience to verify, the other invites the audience in. Both raise the return on work that was checked by a human being.

Reflections Which audiences in your market get the least support in verifying what they read?

IV. On Governance, Identity and Value

Four signals on the new labelling rules, on rights over faces and styles, on the price of visual content, and on the politics of who gains.

22.10 — Europe’s labelling obligations take effect while the transparency code fills up unevenly

From 2 August, the AI Act requires highly realistic AI generated images, video and audio to be clearly labelled as synthetic. Euractiv reports that nearly 190 companies have signed the accompanying voluntary transparency code, including Anthropic, OpenAI, Google, Meta and Mistral, while large Chinese developers with significant European usage, among them DeepSeek and Alibaba, are absent. Lenovo is the only Chinese signatory.

Platform practice is developing alongside the law. TikTok has announced expanded AI transparency tools across the Middle East and North Africa, reporting more than three billion AI generated videos labelled and a seat on the steering committee of C2PA, the provenance standards body.

A labelling regime works to the extent that labels reach the material people actually see. Uneven adoption leaves visible openings, and provenance standards are the practical instrument for closing them.

Reflections How should a newsroom handle synthetic material arriving from a developer who has not signed the transparency code and whose output carries no provenance data?

22.11 — Faces and writing styles become licensable assets

Rest of World reports that Chinese platforms now pay people between 15 and 700 dollars to license their likeness for AI generated content, with catalogues sorted by age, gender and character type. More than 95 percent of the 128,000 microdramas released in China in the first quarter used AI in production, and ByteDance has removed more than 85,000 videos involving unauthorised reproduction of faces and voices since January. In March the Beijing Internet Court held that using someone’s likeness in an AI face swap without permission is illegal, even where the image has been altered. A 21 year old model in Shanghai, Xu Fang, says he will never license his face, because he cannot know how it might be changed.

A similar boundary is being drawn around writing. Unwire reports that ChatGPT has begun refusing direct requests to reproduce the style of living authors, offering a similar feeling instead, a change linked to the authors’ class action before Judge Sidney Stein in New York. Gemini reportedly complies with such requests, Perplexity refuses, and Claude and Copilot comply with a warning.

The lawyer quoted by Rest of World identifies the weak point exactly: licensing terms so vague that people cannot know who will use their likeness or for what. Consent without scope is a signature without a contract.

Reflections What scope, duration and revocation terms would make a likeness licence acceptable for a journalist who appears on camera?

22.12 — Licensing and automation set two different prices for visual content

Reuters has consolidated its visual licensing, announcing that Reuters Connect is now the single destination for licensing news video and pictures, bringing the Reuters picture collection, Imagn sports imagery and the Screenocean archive into one transactional platform. Alphonse Hardel points to rising demand for factual news visuals from publishers, producers, agencies and creators.

In the same week, Futuri launched a platform claiming that newsrooms cut production time by 80 percent and station costs by more than 15 percent a year, turning a 45 minute press conference into publish ready clips in six minutes and one newscast into 45 social assets, with editors keeping approval rights.

Two prices are being set at once. One is for verified, rights cleared material with a provenance chain. The other is for volume produced from material a newsroom already owns. Both belong in a media business, and the interesting question concerns which of the two a customer thinks it is buying.

Reflections Which parts of your visual output gain value from rights clearance, and which gain value from speed?

22.13 — The social contract around AI is written in policy

Josie Stewart and Brooke Tanner argue for Brookings that policy rather than public relations will determine Gen Z’s trust in AI. About half of Gen Z uses generative AI weekly, while only 14 percent of adults under thirty expect its effect on society to be positive and 69 percent trust work completed by humans more. Their proposed remedies are concrete: consent and compensation for creative labour, portable benefits and wage insurance, disclosure standards for data centres, and stronger protection for minors. Fast Company Middle East tests one distributive idea, asking whether the Gulf could convert AI wealth into guaranteed income.

The institutional version appears in Noema, where a coalition of researchers calls for reverse alignment, the deliberate redesign of institutions so societies can absorb AI. The essay names three failure modes worth memorising: productivity without prosperity, execution without verification, and capacity without constraint.

Execution without verification belongs on a newsroom wall. In four words it covers the verification tax, the fabrication rate in AI answers and the gaps in labelling.

Reflections Which policy commitments would genuinely change how a sceptical twenty five year old judges an AI product?

Conclusion

Read together, these signals describe a machine economy that runs on human judgement. Someone stayed in Tehran. Someone verified the video before 200 broadcasters used it. Someone in Manizales spent 48 hours learning to check a message before forwarding it. Someone in an editing suite read the machine’s draft and found the error. That work is expensive because it takes time and experience, and it is precisely what makes an automated answer worth trusting.

Value does not follow from scarcity on its own. It follows when organisations can show how their work is used, keep their name attached to it and negotiate with something in hand. Litigation, licensing and product innovation are all part of that, and none of them substitutes for the others.

Which brings us back to the statue. Socrates would probably enjoy a language model enormously, as a tireless interlocutor and an excellent mirror for weak reasoning, and he would still insist that reproducing information differs from knowing what to ask. That part stays with us. It is midsummer, the future is open, and the questions are ours to choose.

For three decades the working rule of the digital economy was simple: if content could be reached, it could be used, and the argument about payment came later, if at all. That rule is being rewritten this month by legislatures and regulators on four continents at almost the same time, each in its own legal language, each arriving at the same conclusion. Access to news, images and creative work is becoming a permissioned transaction with a price attached.

The links this week describe that shift from several angles. National parliaments in Jakarta, Nairobi, Kuala Lumpur, Bangkok and Brasília are writing AI training and news aggregation into their copyright AI statutes. Competition authorities in Abuja, Brasília and Tokyo are examining whether the use of journalism by large search and AI services is an abuse of market position. Canberra is testing whether the threat of taxation can substitute for a functioning bargaining market.

Running underneath all of this is an economic fact that publishers now measure precisely. Search behaviour has moved into AI answers, advertising inventory on the open web is contracting, and referral traffic has stopped working as the connection between audience and revenue. That is the pressure that turns copyright from a specialist file into a matter of national industrial policy.

The rest of the dispatch follows the consequences: how newsrooms are rebuilding their standards, their audience economics and their pricing; how institutions are setting rules for machine-generated content in their own houses; and what it means for information ecosystems when the world converges, without coordination, on the same answer.

I. On copyright becoming national policy

Eight jurisdictions, one direction. Europe has had its framework for some time through the CDSM directive and now the AI Act transparency layer, tracked in detail by the European picture industry’s continuous record of AI and copyright developments; what is new is that the rest of the world is legislating on the same questions, in parallel.

An Atlas figure carries a networked world whose core is a copyright symbol; uncontrolled data flows are refused on one side, while licensed access continues through secured gateways. (Synthetic Image)

21.1 — Abuja turns a publishers’ petition into a competition inquiry

President Bola Tinubu directed the Federal Competition and Consumer Protection Commission on 6 July 2026 to investigate Meta, Alphabet and X over their use of Nigerian media content, following a petition from the Nigerian Press Organisation, as reported from Lagos. The inquiry rests on the Federal Competition and Consumer Protection Act of 2018 and covers market dominance, conduct, and the ingestion of journalistic and broadcast material into generative AI systems. Possible outcomes range from licensing mandates and revenue-sharing rules to fines and structural remedies.

For publishers the significant element is the choice of instrument. Nigeria’s news industry concluded that no single title can negotiate with a global platform on its own, so it asked the state to address a market failure through competition law rather than waiting for copyright reform. The commission has form here; its 2025 decision imposing a 220 million dollar fine on Meta remains under appeal.

The pattern to note is the pairing of copyright claims with competition powers. Where a copyright statute takes years to amend, a competition authority can open a file in a week, and that difference in speed is shaping which door publishers knock on first.

Reflections

When a national press body asks the state to negotiate on its behalf, what mechanisms are needed to safeguard the independence that same press body defends elsewhere?

Which remedy serves publishers better over ten years, a licensing mandate imposed by a regulator or a price discovered in commercial negotiation?

21.2 — Jakarta writes payment for news into copyright law

Indonesia’s House of Representatives has completed a draft bill amending the 2014 copyright law that would require technology platforms to compensate publishers for aggregating, republishing and link-previewing news content, and for using it to train AI models, according to reporting from Jakarta. Royalties would flow through state-supervised collective management organisations, which would distribute proceeds to news publishers. The draft also grants protection to AI-assisted works only where human involvement is present, prohibits the imitation of an individual’s distinctive style, and requires disclosure of AI use in content.

The collective management route is the interesting design choice. It gives smaller Indonesian publishers a seat at a table they could never reach individually, and it gives platforms a single counterparty; the price of that convenience is state supervision of a revenue stream that funds independent journalism.

Indonesia is combining three separate European debates, the press publishers’ right, the training exception and AI transparency, into one statute. Parliament is in recess until 13 August, and deliberations continue in the meantime.

Reflections

How to maintain state support in the collection and distribution of press royalties compatible with editorial independence?

What is the effect of legislating aggregation, training and authorship in a single instrument?

21.3 — Nairobi rebuilds its copyright regime for the digital economy

Kenya has published the Copyright and Related Rights Bill 2026, the first full overhaul of its copyright regime since 2001, following a consultation opened by the Kenya Copyright Board in March, summarised by intellectual property practitioners in the region. The bill addresses authorship and ownership in the digital environment, including AI-generated works, and introduces notice and takedown procedures, intermediary safe harbours, a specialised copyright tribunal, tighter regulation of collective management organisations, private copying levies and a resale right.

The institutional layer deserves attention. A dedicated tribunal and stronger supervision of collecting societies determine whether rights are enforceable in practice, and enforceability is what converts a legal entitlement into revenue for a newsroom or a photographer.

Kenya is building the plumbing that many jurisdictions skipped when they legislated on AI first and enforcement later. Rights without a forum to assert them remain a statement of principle.

Reflections

Where should a country place its scarce enforcement capacity when both piracy and AI training compete for attention?

Which comes first for a creative economy, a modern statute or a tribunal that can apply it within a commercially useful timeframe?

21.4 — Kuala Lumpur proposes terms for AI use of protected works

Malaysia’s intellectual property office has put a wide set of amendments to the Copyright Act 1987 out for consultation, covering AI, online piracy, orphan works, artists’ resale rights and dispute resolution, as detailed in the national press. The AI element proposes a framework governing the use of copyrighted works by AI systems built on transparency, fairness and appropriate compensation. Procedural changes matter too: either party could refer a royalty dispute to the Copyright Tribunal without the other’s consent, and cases would be resolved within sixty days of the close of hearings.

The office summarised the industry position in one sentence worth keeping: creators, publishers and media want to be asked first, paid fairly and told clearly how their works are being used. That is a workable specification for a licensing market, and it maps closely onto what European rightsholders have asked for since the CDSM directive.

The sixty-day tribunal deadline is the provision with the most commercial weight. Speed of adjudication sets the bargaining position of every party in the room.

Reflections

What changes in negotiation behaviour once either side can force a royalty dispute to a decision within sixty days?

How should a mid-sized publisher price consent when transparency about model training remains partial?

21.5 — Bangkok writes training data into its AI Act

Thailand expects to complete the draft of its AI Act within the current fiscal year, with Digital Economy and Society Minister Chaichanok Chidchob confirming the timeline at AI Governance Week in Bangkok, reported at the start of July. A committee of intellectual property specialists, legal experts and judges convened by the national development agency is defining boundaries for AI training data and examining revenue-sharing mechanisms for rightsholders, with its recommendations to be folded into the final draft. Regulators are separately weighing mandatory labelling of AI-generated content, while officials repeat that enforcement should concentrate on high-risk systems and avoid discouraging foreign investment.

Thailand is doing something few jurisdictions attempt: designing the copyright remuneration mechanism inside the AI statute rather than in a separate reform. That reduces the gap between the transparency obligations placed on model developers and the payment obligations owed to rightsholders.

The investment argument is present in every capital that legislates on AI. Thailand has answered it by concentrating enforcement on high-risk applications while keeping the training data question on the table.

Reflections

Which produces a more durable market, remuneration rules written into an AI act or a separate copyright reform negotiated with rightsholders?

How much regulatory certainty is an AI platform willing to pay for, and at what point does that calculation change?

21.6 — Canberra tests whether the threat of a levy is enough

Australia’s News Bargaining Incentive would charge Meta, Google and TikTok 2.25 percent of their Australian revenue unless they conclude commercial agreements with local publishers, an effective rate falling to around 1.5 percent with sufficient deals and worth an estimated 200 to 250 million Australian dollars a year for journalism, as set out when the draft legislation appeared in April. AI services were explicitly excluded from scope, with the assistant treasurer pointing to separate copyright work led by the Attorney-General. The government then missed its own commitment to introduce the bill before the winter recess that began on 2 July, with parliament returning on 11 August and the distribution mechanism still unresolved.

The design problem is instructive. The 2021 code worked as deterrence because designation was narrow and rarely triggered, and it stopped working when a platform chose to exit the category altogether; the incentive replaces precision with breadth by charging revenue whether or not a platform carries news. News Corp Australia called the delay deeply disappointing, while Nine’s chief executive said he would still prefer platforms at the negotiating table over a tax.

Australia is the live experiment in whether fiscal pressure can manufacture a bargaining market. Every other jurisdiction reading these files is watching how the distribution question is settled, because money that reaches consolidated revenue rather than newsrooms proves nothing.

Reflections

What signal does a missed legislative deadline send to a counterparty whose entire strategy is to wait?

On which basis should a national journalism fund be distributed if the goal is capacity rather than compensation?

21.7 — Brasília chooses statute while its competition authority tests doctrine

Brazil is running both instruments at once. The Senate text of PL 2338/2023, the country’s AI framework bill, requires prior authorisation from and payment to rightsholders before protected works are used for training, a consent-and-pay rule closer to licensing than to any broad exception, while the Chamber of Deputies weighs a text and data mining carve-out that more than fifty industry bodies and the national royalty collection office asked it to reject in May, as traced in a detailed legislative account. In parallel, the competition authority CADE broadened a proceeding against Google in April to cover artificial intelligence, and Google has now filed a 108-page defence asking for the case to be closed, reported in São Paulo.

Google’s central argument is jurisdictional: any obligation of monetary compensation must be established through legislation approved by Congress rather than through a competition finding of exploitative abuse. The company points to France, Australia, Canada and Germany as jurisdictions that created payment mechanisms by statute, notes that publishers retain control through robots.txt, noindex, nosnippet and Google-Extended, and cites data indicating that less than 30 percent of Brazilian publishers’ traffic originates in search.

Brazil therefore offers the clearest illustration of the equilibrium being sought worldwide. Claims and proceedings establish where the legal boundary runs, legislation defines the obligation, and negotiation discovers the price; a workable business model emerges from the combination rather than from any one of them.

Reflections

Where should the line fall between competition law and legislative choice when a market failure is obvious to one side and bluntly denied by the other?

If technical opt-out tools are the answer to compulsory use, what makes an opt-out meaningful for a publisher whose readers are already inside the answer engine?

21.8 — Tokyo studies the market while a licensing platform opens for business

Japan’s competition authority has begun surveying news publishers to establish how Google, OpenAI and other providers of AI-powered news search and summary services use their content, according to a report from Tokyo on 8 July. In the same week, SoftBank started operating a trial version of GaranAI, a platform that acquires article data from news organisations, converts it into a form that is difficult to restore to the original, processes it for machine use and sells it to AI developers, with Mainichi, Sankei, Kyodo and several regional papers supplying content for payment. The official version is planned for January or later, with an intention to extend the model to other categories of copyrighted work.

This is the market-side answer to the same question the legislatures are addressing. A collective supply platform lowers transaction costs on both sides, gives smaller regional titles a route into the licensing market and creates an auditable record of what was supplied and to whom.

Japan is running the regulatory inquiry and the commercial infrastructure in parallel, which is arguably the most practical sequence available. Regulation defines the obligation; an intermediary makes compliance cheap enough to be used.

Reflections

What is the fair split when an intermediary, rather than the publisher, captures the margin on licensed news data?

Which assets should a news agency license through a shared platform, and which are valuable enough to license only directly?

II. On the economics of AI answers

Two signals on where the audience has gone and what remains measurable.

21.9 — Google builds the answer into the front door

Google’s AI Mode replaces lists of hyperlinks with conversational responses written by Gemini, and the consequences are now measurable across the web, as examined in a detailed account of the company’s shift. Users write queries three times as long as before and spend considerably longer inside the interface; one study found that in around 75 percent of sessions users never left AI Mode for the web, a figure Google disputes on methodology. Cloudflare reports that more than half of web traffic is now non-human and that human visits to finance, publishing and retail sites fell close to 40 percent between June 2025 and April 2026, while Wikipedia recorded an 8 percent decline in human visitors.

The advertising consequence follows directly. Benchmarking data covering roughly 20 billion impressions across premium publishers shows publisher ad request volumes down 32 to 37 percent in the United States and 39 to 41 percent in the United Kingdom year on year in the second quarter of 2026, with the operator of that data describing platforms as providing content in situ rather than redirecting. Higher prices per thousand impressions are softening the revenue impact for now, while the underlying inventory contracts.

Britain’s competition authority has already required attribution links and an opt-out from AI summaries, with changes applied globally. That regulatory precedent, arriving while advertising supply contracts, explains why publishers everywhere have moved the conversation from traffic to terms.

Reflections

If inventory contracts by a third while prices rise, how long can a display-funded newsroom treat the revenue line as stable?

What obligations should attach to an interface that answers the question, when the answer is assembled from work someone else financed?

21.10 — Visibility replaces referral as the working metric

New measurements suggest publishers are treating answer engines as a distribution layer rather than a source of clicks, set out with the underlying data by Digiday. AI agent traffic grew 45 percent in the second quarter to 17.7 billion requests across one security network, Meta’s crawlers now generate the majority of it, and ChatGPT accounts for upwards of 80 percent of what little AI referral traffic exists. Publishers are responding defensively: 56.4 percent block at least one AI crawler in robots.txt against 10.3 percent of the open web, while adoption of the llms.txt convention among publishers remains at 3.2 percent and 97 percent of those files received no requests at all in May.

What appears inside the answers matters more than the residual clicks. Analysis of around 300 million monthly searches found Facebook cited as a source in 19.5 million AI Overviews, with Instagram referenced 877,000 times and roughly one search in fifteen answered using social media; the research firm’s chief executive observed that AI search is changing what counts as a source of truth.

That last point deserves to be read carefully by every newsroom leader. When a community post outranks a verified report as a citation, the competition is about the machine’s definition of authority, and that definition is currently set by whoever supplies the most accessible signal.

Reflections

How does a press agency demonstrate its authority to a system that treats a community post and a verified dispatch as comparable evidence?

Which metric should editorial management apply in 2027, if clicks describe an ever smaller part of the relationship?

III. On newsroom strategy and standards

Four signals on how news organisations are responding with governance, product and pricing.

21.11 — Galloni asks competitors to share what they learn about AI

Reuters editor-in-chief Alessandra Galloni used the Andrew Olle Lecture in Sydney on 21 July, as the first representative of a news agency to deliver it, to argue that news organisations should cooperate on AI in the way they already cooperate on physical safety in conflict zones, with the full lecture published by the ABC. She cited the Reuters Institute finding that 10 percent of news consumers globally now use AI chatbots for news, rising to 16 percent among the under-35s, and set out the licensing principles Reuters applies: accurately represented, properly attributed, fairly compensated. Her appeal to Australian competitors was to pool what each newsroom learns about detecting synthetic material and governing AI tools, reported alongside responses from the industry leaders present.

The commercial context in the room was visible. Nine had announced 30 editorial job reductions the same day, News Corp Australasia’s executive chairman described the collection of content by AI companies as the reason they now want legitimate access, and the ABC’s managing director noted the delicate balance in doing deals at all.

The strength of this argument comes from its institutional character. A quality newsroom detects a synthetic voice on a source call because of systems, verification routines and governance committees, and those are precisely the assets that improve when shared across the industry.

Reflections

Which categories of AI knowledge should a news organisation share with its competitors, and which remain a legitimate commercial advantage?

How does an agency adapt its business model for content creation when its clients negotiate licences with AI platforms?

21.12 — AP defines what AI may do in the newsroom

The Associated Press has released updated newsroom standards for artificial intelligence that specify permitted tasks and reaffirm where responsibility rests, published this week. AI may support early-stage research and document summarisation, transcription and translation, suggested headlines, summaries and shotlists, and assistance with grammar and search optimisation, with every output reviewed and edited by AP journalists before publication. Generative AI remains prohibited for creating, altering or enhancing news photography, new guidance covers the verification and reporting of AI-generated and manipulated material, disclosure is required where generative AI plays a material role, and the standards now extend to AI coding assistants in software development.

The extension to coding assistants is the part most newsroom policies omit. Editorial standards traditionally stop at the published article, while a growing share of what audiences experience is produced by systems that engineers write with machine assistance.

Standards documents of this kind are becoming the working currency of trust between agencies, publishers and AI developers. A licence is easier to negotiate when both parties can point to a written account of how the content was produced and verified.

Reflections

Where does the responsibility for an error introduced by an AI coding assistant belong inside a newsroom’s accountability structure?

What level of AI involvement is material enough to disclose, and who in the organisation decides?

21.13 — Publishers rebuild the business around habit

Three pieces this week describe the same strategic answer to declining referral traffic. Financial Times chief executive Jon Slade describes bringing products, subscriptions, events and commercial services into one connected customer relationship, argues that human judgement remains the organisation’s greatest differentiator in a world of abundant information, and explains the FT’s role as a founding member of the SPUR coalition, which aims to establish common standards for licensing journalism to AI developers. Viafoura’s chief executive makes the measurement case that frequency is the metric that survives the AI era, since a reader who arrives twelve times a month by choice is a customer while a monthly arrival from search is a rounding error. And the format that builds that habit is increasingly the explainer, now treated as a core audience growth strategy by the FT, Bloomberg and Vox Media, with the FT appointing a dedicated explainer editor and Bloomberg using the format to build trust and routine.

The economics are straightforward. Reach that arrives through an intermediary can be reallocated by that intermediary at any time, while frequency built on a direct relationship is an asset the publisher owns and can price.

There is a caution in the same material. AI handles straightforward explanation competently, so the defensible version of the explainer carries reporting, judgement and a recognisable human voice; the signal in the noise has become the product.

Reflections

What frequency threshold should a publisher treat as the boundary between an audience and a customer base?

How does a news agency measure habit when its content reaches the public through someone else’s brand?

21.14 — Algorithmic pricing surfaces in renewal emails

Subscribers to Wired, NJ.com and The Wall Street Journal have begun receiving renewal notices carrying the line that the price was set by an algorithm using their personal data, documented with several examples. One NJ.com reader was quoted 130 dollars a year while others in the same conversation reported 145 and 175; a Wired subscriber was offered 40 dollars against a list rate of 80. The disclosure is required by New York’s Algorithmic Pricing Disclosure Act, in force since January, and Hearst confirmed it applies dynamic pricing across all of its newspaper markets at renewal, describing a long-standing approach based on tenure and, more recently, digital engagement.

The practice itself is ordinary revenue management; the disclosure changes its character by making differential treatment visible between readers who talk to each other. One subscriber cancelled, a class action has been filed against another publisher for failing to disclose the practice, and a consumer advocate warned about access to essential information priced by personal characteristics.

For an industry whose core proposition is transparency, the governance question answers itself. Personalised pricing requires a policy that a newsroom would be comfortable reporting on if a competitor were doing it.

Reflections

Which subscriber attributes are legitimate inputs to a price, and which would damage the relationship if the reader discovered them?

How should a publisher explain differential pricing to its own audience before a regulator explains it for them?

IV. On institutions setting rules for themselves

Two signals on enforcement capacity and on how public bodies handle machine-generated content in their own work.

21.15 — A Cyprus study exposes the enforcement gap in EU political ad rules

Research on the 2026 Cypriot parliamentary election found substantial weaknesses in the enforcement of the EU’s Transparency and Targeting of Political Advertising regulation, in force since last October, analysed for Tech Policy Press by Liz Carolan. Researchers at the Mediterranean Digital Media Observatory recorded at least 2,457 political ads passing Meta’s checks, with fewer than one percent removed while active and one candidate continuing to advertise after nineteen recorded violations; TikTok, which has banned political advertising globally since 2019, carried 887 political ads from 67 advertisers, with a single removal in the first 220 days of the regulation. Google’s account-level identity verification materially constrained entry, with one advertiser found in breach. Following meetings with the researchers, 687 takedown notices appeared on Meta within 30 hours.

The resource asymmetry is the finding that should concern European institutions most. Two researchers working for about a month each, funded by grants and university salaries, produced enforcement outcomes that platform systems had missed across seven months, ahead of elections in France, Spain, Italy and Poland.

An unenforced transparency rule creates an opening for exactly the coordinated interference it was written to prevent. Regulation without inspection capacity functions as a statement of intent.

Reflections

Who should carry the cost of verifying platform compliance, and what happens to European elections while that question remains open?

What would effective independent audit capacity for political advertising cost?

21.16 — Legislators write guardrails for their own use of AI

The European Parliament is testing EPGenAI Hub, an in-house platform giving members and staff access to Llama, GPT-OSS and Mistral Small running inside Parliament systems alongside externally hosted ChatGPT and Claude Sonnet, with a possible rollout from September, reported this month. Around 2,100 staff, roughly a fifth of the total, already use AI daily; staff rules from April 2024 require labelling of AI-assisted work, while members face no obligation to disclose whether an amendment, question or speech was drafted with AI. Committee chairs have asked for guidelines, with one member arguing that Parliament should lead by example as the AI Act’s transparency rules become applicable, and the institution has also moved search to Qwant and disabled built-in AI on members’ tablets.

South Korea has taken the same question into statute. Follow-up decrees to its AI Basic Act took effect on 22 July, making watermarks mandatory on generative AI output with fines up to 30 million won, activating a management regime for high-impact systems and adding a certification scheme that eases public procurement, as reported by Yonhap News TV. A grace period of at least one year acknowledges that large firms such as Naver and Kakao have compliance structures while smaller developers do not.

Two institutions, one principle: the bodies writing the rules for machine-generated content are applying them to their own output first. That is how transparency obligations acquire credibility with the public and with the market.

Reflections

What should voters be entitled to know about the role of AI in the drafting of the laws that bind them?

How long can a labelling obligation remain credible if the tools to verify the label are held by the same companies producing the content?

Conclusion

Read separately, this week’s files look like ordinary national politics: a competition inquiry in Abuja, a draft bill in Jakarta, a tribunal in Kuala Lumpur, a delayed levy in Canberra, a defence filing in Brasília, a survey in Tokyo. Read together, they describe something that has happened rarely in the history of copyright, which is a set of legislatures on four continents arriving at the same question within the same quarter, without a treaty, a coordinating body or a common legal tradition.

The question they share is who bears the cost of producing verified information and who captures its value once a machine has processed it. Every answer on the table this week accepts the same economic premise: quality journalism, photography and creative work are scarce, they are expensive to produce, and free access to them was always a transfer of value rather than an absence of one. There is no free lunch, and the world is now writing the invoice into national law.

What follows is a distributed licensing market with local prices, local collecting arrangements and local enforcement, resembling the way music rights developed across jurisdictions over decades, compressed here into a few years. That creates genuine complexity for any organisation operating across borders, and it creates real leverage for the first time since the open web became the default distribution layer.

The Atlas figure holding a world with a copyright symbol at its core captures the position accurately, including its cost. Carrying that world means maintaining the legal, technical and commercial infrastructure that makes consent verifiable and payment collectable, in every market, permanently.

For news organisations the practical instruction is to prepare for a market in which rights are enforceable and prices are negotiated rather than assumed. For press agencies, whose output functions as critical information infrastructure for an entire ecosystem, the additional task is to ensure that the emerging licensing architecture recognises wholesale supply as well as consumer-facing publishing. For regulators the test is capacity; a rule that reaches the statute book and stops there teaches the market exactly what it can afford to ignore.

This week the same company keeps appearing in the frame. Google shows up in a search setting that trains its models, in a regulatory review in Tokyo, in a copyright rewrite in Jakarta, in the licensing deals that fund Reddit, and in the traffic that publishers are now openly threatening to withhold. No single actor is the story, yet one large gateway sits at the centre of almost every signal, and that concentration is worth watching on its own terms.

Around that centre, three forces are moving at once. Newsrooms are learning to work with AI in practical, supervised ways, from print layout in Barcelona to broadcast training in Delhi. Regulators in Europe and Asia are testing how far consent, competition law and copyright can be stretched to cover machine reading of published work. And publishers are building new products to keep the value of their journalism inside their own walls, while a few experiment with selling that work directly into the machine.

The connecting question is an economic one. Quality reporting is scarce and expensive to produce, and the current arrangement lets a small number of platforms read it, summarise it, and answer the reader before the reader ever reaches the source. This dispatch traces how newsrooms, lawmakers and publishers are each trying to restore a workable price for that scarce content, and why so many of the roads lead back to Google.

I. On the working newsroom

The first cluster of signals is about craft: how editorial teams are absorbing AI into daily practice without surrendering judgement.

20.1 — Barcelona keeps the human in the print edition

The Spanish sports daily SPORT is testing a project called ESCRIBA, which explores how AI can assist the layout of its print edition, helping with selection, hierarchy and adjustment while professionals keep final control. The initiative is funded through Spanish grants for integrating AI into media value chains, drawing on European NextGenerationEU money. It treats AI as support for the print process rather than a replacement for the editors who shape the paper.

For a publisher running an intense daily news cycle, the appeal is operational: reduce friction in the precision-heavy final hours before the edition closes, and let staff concentrate on judgement rather than mechanics. Print still matters commercially for many titles, and the case for AI here is efficiency inside a format that already earns real revenue. The stated safeguard is human supervision at every stage that carries editorial responsibility.

The wider pattern is a sober one. The most durable AI adoption in journalism is arriving in the unglamorous parts of the workflow, where it saves time without touching the reader’s trust in what is reported.

Reflections Which back-office tasks can a newsroom automate before readers begin to sense a change in the product itself? How should a publisher document human oversight so that “a professional decided this” remains verifiable rather than assumed?

20.2 — OpenAI and public broadcasters train the people, not only the tools

Two training efforts landed in the same week. OpenAI published a ChatGPT 101 for Journalists collection inside its Academy for News Organizations, with practical lessons on prompting, reusable Skills and workflow Agents, all framed around approved sources and human review. In India, the public broadcaster Prasar Bharati ran a three-day programme for Doordarshan and All India Radio staff, covering generative AI, prompt engineering, responsible-AI practice and a long list of tools for scripting, translation and production.

The strategic point is that skills, culture and mindset are the real constraints on adoption, more than the technology itself. Training the workforce, whether at a commercial vendor’s academy or inside a state broadcaster, is how organisations turn a policy document into daily practice. It also decides who sets the terms of that practice, the platform or the publisher.

A newsroom that only buys tools without building internal capability will find its habits shaped by whoever wrote the training material.

Reflections When a technology provider also trains your journalists, how do you keep your own editorial standards at the centre of the lesson? What is the right balance between vendor-led instruction and independent, cross-newsroom skill building?

20.3 — Newsroom leaders converge on a short list of principles

The Columbia Journalism Review assembled newsroom leaders and policy experts to explain how to develop AI guidelines that actually hold. The recurring advice is that a written policy is only the start; it needs training, oversight and regular revision to stay alive. Editors from Bloomberg, the Wall Street Journal, Reuters, Axios and others describe a common core: transparency, human oversight, accountability and authenticity, with generative writing of full stories largely off-limits.

For decision-makers, the useful lesson is procedural. The Wall Street Journal treats its guidelines as a living document reviewed every six months, Reuters runs a governance committee with a kill switch chaired by its editor in chief, and several outlets deliberately include AI skeptics alongside enthusiasts. Policy works when it is editorially driven and revisited, and fails when it becomes a PDF nobody reads.

The through-line across all these newsrooms is that accountability stays with a named human for every published word, whatever the tool in the pipeline.

Reflections How often should an AI policy be revised before “living document” becomes an excuse for having no settled rules at all? Who in your organisation holds the authority to stop an AI workflow the moment it produces something unpublishable?

20.4 — Tasaka argues that agentic AI changes the resource question

In a column for Editor & Publisher, media technologist Guy Tasaka contends that agentic AI will save local media in a way generative chatbots never could. His argument is that a chatbot answers a single request, while an agent can accept an open-ended goal and work through many steps on its own, planning, using tools, checking its work and reporting back. For newsrooms hollowed out by decades of budget pressure, he frames this as staff they could never afford, available at low cost.

The commercial implication is real for small publishers with scarce resources: work that was previously impossible for lack of hands becomes attempt-able. Tasaka points to coding agents such as Claude Code, Codex and Cursor, and to open-source projects worth watching, while a skeptical reader comment on the same page presses him on what any of this concretely does for a local newsroom’s actual problems. That exchange captures the honest state of agentic AI in media: promising in principle, still unproven at the level of daily local coverage.

The prudent reading treats agentic AI as an early-stage capability to pilot carefully, with the same human accountability that governs every other tool in the newsroom.

Reflections Which local-media task is well enough defined that you would trust an autonomous agent to attempt it end to end? How do you measure whether an agent has genuinely added capacity, rather than shifting the work of supervision onto already stretched editors?

II. On law in search of a price

The second cluster is regulatory. Across three jurisdictions, lawmakers and enforcers are testing how existing rules apply when machines read published work, and each is groping toward a workable price for that use.

20.5 — The EDPB closes the door on consent as a scraping excuse

The European Data Protection Board adopted Guidelines 03/2026 on web scraping for generative AI, concluding that consent will rarely be a workable legal basis for large-scale collection, since firms scraping the open web have no direct relationship with the people whose data they gather. The guidance is explicit that publishing something on an openly accessible page does not amount to consent to have it scraped, and that the absence of a robots.txt file is not consent either. That leaves legitimate interest under Article 6(1)(f) as the basis the Board expects most private entities to invoke, subject to a three-part test and a careful balancing exercise.

For publishers and platforms the practical consequences are concrete. The guidance favours narrower collection, records of scraped sources, opt-out mechanisms and public transparency, and it stresses that once a model is trained, personal data cannot easily be removed from it. The document is open for consultation until 30 October 2026, so its final shape is still in motion.

Under European data protection law, the burden is shifting toward those who read the web at scale to justify and document that reading, rather than assuming the open web is a free input.

Reflections If consent cannot cover open-web collection, what standard of transparency should the public reasonably expect from any firm training on published material? How should a publisher signal its terms so clearly that a later “we reasonably expected access” defence becomes untenable?

20.6 — Tokyo and Jakarta test competition law and copyright on the same problem

Two Asian jurisdictions moved in the same week. Japan’s Fair Trade Commission opened a review of AI search use of news content, examining zero-click summaries and whether the practices of Google and LY Corp breach the Antimonopoly Act through abuse of a dominant position, with some 370 news organisations engaged. Indonesia, meanwhile, is preparing a copyright rewrite that would require platforms to compensate publishers for aggregating, previewing or training on news content, route payments through collective management organisations, and could revoke a platform’s local operating licence for non-compliance.

The strategic significance is that two different legal instruments, competition law in Tokyo and copyright in Jakarta, are converging on the same underlying question of who pays for machine access to journalism. Google has already criticised the Indonesian draft as overbroad and warned it could leave the country an outlier, while the government frames unregulated generative AI as a threat to human creation. From a European vantage point, the natural comparison is the droit d’auteur and CDSM tradition rather than any American doctrine, and the outcomes abroad remain genuinely open.

Claims in court and negotiations at the table are two halves of one process; litigation sets the legal boundaries while licensing settles a workable price, and neither alone produces a stable model.

Reflections When competition law and copyright law target the same conduct, which is more likely to produce a durable payment mechanism for publishers? How should a news agency position itself when the compensation flows through a state-supervised collecting body rather than a direct commercial deal?

20.7 — UK publishers get a diagnostic for their own AI position

The Professional Publishers Association released a Publisher AI Licensing Framework, a practical diagnostic that maps variables such as size, content vertical and revenue model to show which business models are most exposed to AI and which factors offer resilience. It is member-facing and built to help each publisher assess its own position and choose a strategic pathway rather than prescribing a single answer.

The value for management is that it turns an anxious, anecdote-driven debate into a structured self-assessment. A specialist title with scarce, defensible content sits in a very different position from a commodity-news site, and a framework that names those differences helps a board decide whether to license, withhold or build. It treats licensing as a strategic choice grounded in the specific economics of each publisher.

The lesson is that resilience in the AI era depends less on the technology than on how scarce and defensible a publisher’s content actually is.

Reflections Which parts of your catalogue are scarce enough that an AI developer would rather license them than do without? How do you price content for machine reading when you still cannot fully observe how that content is used inside a trained model?

III. On the gateway and its traffic

The third cluster brings the platform question into focus. Here the dependence on a single search gateway becomes explicit, and publishers begin to test their leverage.

20.8 — Google turns search interactions into training material

Google is rolling out a Search Services History setting that can use searches, including queried files, uploaded images, voice searches and Search Live recordings, to train its AI, with a related Save Media option that stores media from search interactions to improve its models. The change extends the flow of training data from the open web into the search interaction itself.

For strategists the implication is about the scale and integration of one company’s data position. The same interactions that answer a user’s question also feed the models that will answer the next one, which deepens the advantage of the largest gateway. This is the kind of concentration that deserves attention as an early-warning signal about market structure, treating the platform as an object of analysis rather than a culprit.

The reach of a dominant search service now includes not only what the world publishes, but how the world searches.

Reflections What should a publisher assume about the reuse of any content or query that passes through a dominant search interface? How far can one company integrate search, data and model training before the market itself needs a corrective?

20.9 — Publishers start threatening to walk away from Google

Big publishers are now openly weighing whether to leave Google Search, calculating whether the traffic Google still sends is worth feeding the AI features that answer readers without a click, with reports of at least one major title’s chief executive threatening to delist unless compensation talks progress. The same anxiety is quantified by Arc XP’s citation of the Reuters Institute finding that only 4 percent of AI chatbot users say they often click through to original sources, against 19 percent from search. Smaller sites, the commentary notes, cannot credibly threaten Google and should instead reduce dependence on traffic they do not control.

The strategic reality is that the old exchange, content for referral traffic, is being renegotiated in public, and the balance of power depends on how much direct audience a publisher owns. Visibility is shifting from clicks to citations, which rewards outlets that AI surfaces read, trust and name. Owned assets such as email lists, subscriptions and direct reputation are becoming the real defence.

The publishers with the most leverage are those least dependent on the gateway they are threatening to leave.

Reflections At what point does the traffic a platform sends cost a publisher more in lost value than it delivers in audience? How does a news organisation build enough direct relationship with readers to make “leaving Google” a credible option rather than a bluff?

20.10 — Reddit debates whether feeding the machine undercuts its own business

Reddit is openly debating whether its content-licensing deals with OpenAI and Google compete with the advertising business built on the same community data. Its ads chief argues the two draw on different layers, licensing raw content while ad targeting uses engagement signals the company keeps to itself, though he concedes the question is unsettled for a young platform. Analysts add a second worry already in Reddit’s own risk disclosures: AI answer engines could reduce the organic traffic its ad business depends on, even as licensing fees rise.

The lesson for any content owner is that selling data to train a model and selling access to an audience can pull against each other. A publisher’s data can be both its product and its competitor’s raw material, and the terms of a licensing deal decide which effect dominates. Reddit’s internal debate is a useful preview of a choice many publishers will face.

Every licensing decision carries a quiet trade between short-term revenue and the long-term value of the data itself.

Reflections How does a content owner license its data for AI training without eroding the very signal that makes its audience valuable to advertisers? Which contractual limits, on identification, targeting or disintermediation, matter most when selling access to your community’s content?

IV. On building value inside the walls

The final cluster is about strategy: how publishers are trying to keep the value of their journalism close, and what happens when someone treats reporting purely as raw material.

20.11 — The Washington Post builds an answer layer publishers can own

The Washington Post’s technology arm, Arc XP, launched Ask The News, an AI answer layer that publishers embed on their own sites to answer reader questions using their own reporting, with attribution, editorial guardrails and a refusal to answer where the source reporting is thin. It bundles a subscription gateway that meters answers rather than articles, and contextual advertising, so the publisher keeps the interaction, the intent data and the revenue.

The strategic logic is to reclaim the reader conversation that platforms have been capturing, and to do it on the publisher’s own property. Rather than surrender the question-and-answer moment to an external chatbot, the publisher hosts it, keeps the data and monetises it directly. It is a constructive response to the click-through collapse the same company documents.

The design principle is simple: keep the value of the journalism inside the walls that paid to produce it.

Reflections What does a publisher gain, and give up, by turning its journalism into a conversational answer service rather than a set of articles? How do you keep an owned answer layer honest about the limits of your own reporting when the commercial incentive is to always have an answer?

20.12 — The New York Times treats local outlets as partners, not prey

The New York Times is launching The Local, a community journalism initiative that begins in August with a free three-times-a-week newsletter for Minneapolis and St. Paul. The Times says it wants to reach people who have never engaged with it, and, notably, frames local outlets as partners rather than competitors, promising to promote other newsrooms’ stories and look for ways to collaborate.

For a national institution moving into local territory, the partnership framing is the strategic choice worth watching. It positions the newsletter as additive to a “vibrant media ecosystem” rather than extractive from it, and tests whether a large brand can deepen reader relationships locally without hollowing out the outlets already there. Whether the collaboration proves real or rhetorical will show in how traffic and credit actually flow.

The interesting question is whether scale and locality can be combined in a way that strengthens the wider ecosystem rather than absorbing it.

Reflections When a national brand enters a local market as a self-declared partner, what would genuine partnership look like in the numbers? How should established local outlets respond to a larger player that promises to promote them while competing for the same attention?

20.13 — A start-up treats reporters as an input to AI, and a tribunal becomes a scoreboard

Two signals show what happens when journalism is treated primarily as raw material. A venture-backed start-up profiled by the Washington Post has raised 70 million dollars by hiring local reporters and feeding their work to AI, placing journalists in statehouses as an input to a technology product. Separately, the Peter Thiel-backed venture that once staged AI “trials” of reporters has rebranded as The Primary and now scores and ranks journalists and outlets by a language model’s reading of their rigour, weighing source attribution, tone and right of reply.

The strategic reading is that the value of reporting is being captured and repackaged by intermediaries, whether as training input or as a ranking product, with the journalists further from the resulting revenue. The Primary’s own chief executive says the trigger was watching AI systems compress a reporter’s work while “the machine took the value,” a candid description of the same asymmetry publishers are fighting elsewhere. Both models raise questions about who benefits from journalism once it becomes a feedstock.

When reporting is treated as a scarce raw material, the decisive question is who sets its price and who keeps the margin.

Reflections If investors will pay to turn local reporting into an AI input, what does that reveal about the price publishers should be charging for the same work? Who should hold the authority to score a journalist’s rigour, and what accountability should attach to the scorer?

Read together, these signals describe a single structural problem seen from many angles. Quality journalism is scarce and costly to produce, and a small number of platforms can currently read it, summarise it and answer the reader before the reader reaches the source. The newsroom stories show publishers professionalising their use of AI; the regulatory stories show three jurisdictions trying to attach a price to machine access; the platform and strategy stories show publishers testing their leverage and building products to keep value close. The common thread is the search for a workable equilibrium in which reading has a cost.

Google sits at the centre of that picture, not as a villain, but as the largest gateway through which most of these questions pass. It trains on search interactions, it is the subject of a competition review in Tokyo and a copyright rewrite in Jakarta, it funds and pressures publishers at the same time, and it is the platform that publishers are now willing to threaten to leave. A gateway of that scale carries a social responsibility proportionate to its reach, and treating its behaviour as an early-warning signal is simply prudent market analysis.

For news organisations and press agencies in particular, the strategic imperative is clear enough. Press agencies are critical information infrastructure, and their value lies in scarce, verified, defensible content that machines cannot cheaply reproduce. The publishers with real options this week are those who own their audience, document their standards and can name a price for access to their work. The equilibrium the law is searching for will only hold if publishers themselves treat their journalism as the scarce asset it is, and price it accordingly.

The lesson of the week is an old economic one dressed in new technology. There is no such thing as a free lunch, and the reading of the world’s journalism was never going to stay free forever.

When a crawler takes thousands of pages for every visitor it returns, the arrangement stops being reciprocal and becomes extraction. The links this week describe the question asked in many jurisdictions and many registers.

Who pays for the knowledge that machines now consume, and what happens to the commons when the people who fund it decide to close the gate? Regulators in Paris, Copenhagen, Abuja and Dublin are testing legal answers. Publishers in the United States and the United Kingdom are testing technical ones. Researchers in Oxford are measuring how the same systems reshape opinion while the argument over payment continues.

These threads converge on one structural choice. If the automated reading of the web can no longer be assumed, the architecture of access itself becomes the battleground. Publishers are now willing to build a wall between human readers, who remain welcome, and machine readers, who must ask permission, accept limits and pay. This dispatch follows the themes of licensing, trust, governance and platform power towards that wall, and closes on the image it produces.

Chapter 1 — Licensing and the Search for Equilibrium

19.1 — Getty shows that litigation and licensing are one process
Source
• Getty Images Fought AI In The Courts, Lost, Then Lucratively Licensed. The New Publishing Standard.

Dispatch
Getty Images sued Stability AI in the United Kingdom in 2023, then saw most of its claims fall away, with the High Court rejecting the secondary copyright argument in November 2025 on the ground that model weights did not themselves store copies of its images. Rather than treat that ruling as an ending, Getty separated two distinct products, the right to train a model and the right to display an image, and licensed only the second. It signed a display agreement with Perplexity in October 2025 and a comparable one with OpenAI in June 2026, both conditioned on attribution and linking, and neither granting training rights.

For decision-makers the lesson concerns product design rather than victory in court. Training rights and display rights carry different risks and command different prices, and a publisher can decline the first while selling the second. Getty attached attribution as a contractual term, which turns provenance into a paid feature rather than a favour. The market read this as value creation, with reported share-price increases ranging from roughly ninety to over two hundred per cent across outlets, against a Q1 2026 net loss of 4.4 million dollars, down sharply from a year earlier.
This is what equilibrium looks like when it forms. Claims in court establish the legal boundaries, and negotiated licences establish workable prices, and from that combination a sustainable business model emerges. The News Corp arrangement with OpenAI, reported above 250 million dollars over five years, and comparable deals at Axel Springer, Le Monde and Dotdash Meredith, suggest that the firms doing the rights and metadata work early are negotiating from strength.

Reflections
• How should a news organisation price display access separately from training access without undermining either?
• What does a firm need to prepare in its archive and metadata before it can negotiate from strength rather than from grievance?

19.2 — Denmark and France put a European price on press content
Source
• Denmark intervenes in EU court case over publishers' rights. Reuters.
• French antitrust watchdog orders Meta to resume talks with media groups over publishing fees. The Star / Reuters.
• French watchdog orders Meta back to press payment talks after copyright deals expire. Euronews.

Dispatch
Two European institutions moved in the same week to give press publishers' rights concrete force. Denmark filed a written intervention in the European Court of Justice Streamz case, supporting Belgium against Google, Meta, Spotify, Streamz and Sony, and arguing that a ruling for the platforms would dilute the press publishers' right under Article 15 of the CDSM directive.

In Paris the competition authority found that Meta had likely abused its dominant position by imposing its own method for calculating fees while withholding the information publishers needed to assess them, and ordered Meta to present a payment plan within fifteen days.

For strategists the significance lies in where the pressure originates. The French decision rests on competition law rather than copyright alone, which means the remedy addresses conduct, the refusal to share information and the imposition of terms, rather than the abstract question of whether content was used. The author says the authority deliberately declined to name a provisional sum, to avoid creating a focal point for the negotiation, which leaves the price to the parties while the state enforces the duty to bargain in good faith.

The pattern across Copenhagen and Paris is a European insistence that content
19.4 — Newsrooms write the rules for AI they can defend
Source
• ABC unveils new AI policies. ABC (Australia).
• A new study looks at the skills journalists are losing and gaining because of AI tools. Nieman Lab.

Dispatch
Australia's public broadcaster set out formal rules for the use of AI in its journalism, joining a growing group of newsrooms that treat internal governance as a condition of public trust rather than an afterthought. Alongside this, new research summarised by Nieman Lab examines which skills journalists gain and which they lose as they adopt AI tools, from faster research and transcription to a measurable risk of atrophy in verification and source work.

For newsroom leaders the two items describe a single management problem. A public AI policy protects the institution externally, and a skills strategy protects it internally, since a newsroom that automates its craft without maintaining its judgement erodes the very expertise that makes its output worth licensing. The commercial case for quality journalism and the training case for journalists are the same case.
The broader pattern connects standards to scarcity. Verified, accountable reporting commands a price precisely because it is costly to produce and hard to replicate, and a newsroom that lets its verification skills weaken lowers the value of what it sells. Governance and craft together defend the asset.

Reflections
• Which editorial skills must a newsroom actively preserve as AI absorbs routine tasks, and how would a leader measure their decline?
• How does a published AI policy change the trust relationship with audiences, funders and licensing partners?

19.5 — Authenticity inverts when synthetic feels more real
Source
• AI footage of politicians seen as 'more authentic' than the real thing. The Star.
• Lily Jay faces claims of AI-generated charity videos. Let's Data Science.
• Bombay High Court grants interim relief to Preity Zinta against deepfakes. NDTV.

Dispatch
Three items describe the same disturbance from different angles. Research reported by The Star found that audiences sometimes judge AI-generated footage of politicians to be more authentic than genuine recordings, a charity faced allegations that its emotional appeal videos were synthetic, and the Bombay High Court granted the actor Preity Zinta interim relief against deepfakes using her likeness. Perception, fundraising and personality rights are each affected.

For decision-makers the finding on perceived authenticity carries the sharpest warning. When a synthetic image satisfies an audience's expectation of what truth should look like more fully than reality does, the traditional defence of showing the real recording weakens. Courts are responding with likeness protection, as in the Zinta ruling, and charities and campaigns face a reputational exposure that arrives faster than verification can.
The pattern points back to provenance. Authenticity can no longer be assumed from the look of an image, so it must be established through attribution, chain of custody and verifiable sourcing. This is the same asset that Getty priced in the first chapter, trust made legible, and it explains why display deals condition payment on credit and linking.

Reflections
• When audiences find synthetic footage more convincing than real footage, how does a newsroom prove that authentic material is authentic?
• What responsibility does a charity or campaign carry to disclose AI-generated emotional appeals to its donors?

19.6 — Detection struggles even at the company that built the image
Source
• Meta AI image detector fails to identify some of its own cropped AI images. Reuters.
• How one fake photo let Russian propaganda cast doubt on the Kyiv Lavra strike. Kyiv Independent.
• AI fakes about the Lavra, fake Euronews reports and manipulations about Crimea. Detector Media.

Reuters reported that Meta's own AI image detector failed to flag some AI images it had generated once they were cropped, which shows how easily provenance signals break under ordinary editing. In Ukraine, a single fabricated photograph allowed Russian propaganda to cast doubt on the Lavra strike, and Detector Media catalogued a wider series of AI fakes and impersonated Euronews reports over one week in June. Fabrication and doubt operate together.

For editors and platform operators the failure of first-party detection matters because so much verification policy assumes that machine-generated content can be reliably labelled. When a detector misses cropped images from its own model, watermarking and automated flags become one input among several rather than a settled solution. According to the Ukrainian reporting, a Meta spokesperson attributed erroneous fact-check labels on posts about the Lavra attack to a technical fault, which itself became material for the disinformation campaign.

The broader pattern is that doubt is now a weapon in its own right. A propagandist does not need every fake to be believed, only enough uncertainty that audiences distrust genuine reporting. That raises the value of institutions whose verification can be traced and defended, and it explains why provenance is becoming a paid and contractual feature rather than a technical afterthought.

Reflections
• When a company's own detector cannot reliably identify its own edited outputs, how much weight should regulators place on automated labelling?
• How does a newsroom counter a disinformation strategy whose goal is doubt rather than belief?

Chapter 3 — Platform Power, Governance and Public Discourse

19.7 — AI-mediated communication can steer opinion at scale
Source
• AI-powered social media can subtly manipulate opinion at scale, new study finds. Oxford Internet Institute.
• Misinformation inciting harm to refugees, UN. The Straits Times.

Dispatch
Researchers at the Oxford Internet Institute and the Hasso Plattner Institute found that large language models systematically shifted the position expressed in social media posts on contested topics, even when instructed to preserve the original meaning, and that these small changes could accumulate across millions of interactions to influence public opinion. In one test of X's "Explain this post" feature, the imbalance traced back to a single instruction telling the model to challenge mainstream narratives. Separately, the UN warned that misinformation is inciting real-world harm against refugees.

For policymakers the important finding concerns implementation rather than the models alone. The direction and size of the effect depended on choices made by the platform, which means accountability attaches to product decisions that are documented and auditable. Senior author Sandra Wachter noted that AI-mediated communication is a subtle form of influence that the law has yet to address, and the study observes that neither the EU AI Act nor the Digital Services Act directly covers it.

The pattern links quiet editorial nudges to visible harm. When a rewriting feature tilts contested debate and when misinformation drives violence against vulnerable groups, the governance question moves from content moderation to the design of the tools that shape expression. The largest platforms carry a corresponding responsibility to make those design choices transparent.

Reflections
• When a single platform instruction can tilt opinion on contested topics, what disclosure should regulators require about such prompts?
• Where does responsibility rest when harm follows from a platform's implementation choices rather than from the underlying model?

19.8 — Platform gatekeeping reaches creators and definitions
Source
• YouTube warns UK creators over government's new content control rules. The News.
• The Guardian on Meta's algorithm and digital publishers such as LadBible. The Guardian.
• Committee to Protect Journalists in turmoil after 17-1 vote on definition of journalist. Washington Free Beacon.

Dispatch
Three items show gatekeeping expressed through rules, algorithms and definitions. YouTube warned United Kingdom creators about the effect of new government content-control rules on their reach, The Guardian reported how changes to Meta's algorithm reshaped the fortunes of digital publishers such as LadBible, and the Committee to Protect Journalists faced internal conflict after a 17 to 1 vote over how to define a journalist for its Gaza casualty list. Distribution, reach and professional identity are each subject to a gate.

For strategists the common thread is dependence on a definition or a setting controlled by someone else. A publisher whose audience depends on an algorithm can lose reach without notice, a creator whose livelihood depends on platform rules is exposed to policy shifts, and a profession whose boundaries are contested struggles to defend its members and its standards. Control over the criteria is control over the outcome.

The pattern connects platform power to institutional authority. When the largest gateways set the terms of visibility and when professional bodies divide over who counts as a journalist, the infrastructure of public information becomes a matter of governance rather than of neutral plumbing. The organisations that own the gate carry responsibility for how it is used.

Reflections
• How should a publisher structure its audience relationships to reduce exposure to a single platform's algorithm?
• Who should define who counts as a journalist, and what protections and duties should follow from that definition?

Chapter 4 — Towards the Wall Between Humans and Machines

19.9 — AI Overviews and the measured cost to publishers
Source
• AI Overviews cut publisher clicks 39.8% in first randomized study. PPC Land.
• Journalism job cuts in 2026. Press Gazette.

Dispatch
The first randomised field experiment on Google's AI Overviews, run across 1,065 United States participants and more than 68,000 searches, found that the summaries cut outbound publisher clicks by 39.8 per cent and raised zero-click searches by 34.5 per cent. Because participants were randomly assigned to see or hide the overview, the study provides a causal estimate rather than a correlation, and it observed that more than 87 per cent of overviews sat above every organic result. Press Gazette continued to track journalism job cuts through 2026.

For news organisations this connects a mechanism to a payroll. When a summary answers the query on the results page, the visit that once funded the reporting does not occur, and the causal design here strengthens the case that the loss is real rather than seasonal. The proximity of measured traffic decline and continuing newsroom cuts describes a direct pressure on the economics of production.

The pattern establishes the premise for everything that follows in this chapter. Information is scarce and costly to produce, and a distribution model that consumes it without returning readers removes the revenue that pays for it. Once publishers can measure that loss precisely, the question of whether to keep the gate open becomes a financial calculation rather than an article of faith.

Reflections
• When outbound clicks fall by nearly forty per cent, which revenue models can still sustain original reporting at scale?
• What evidence would persuade a board that opting out of a dominant search engine protects rather than harms the business?

19.10 — Adaptation without walls, from Reuters to Yahoo
Source
• Reuters launches Model Context Protocol server to bring trusted news into AI workflows. Reuters / Editor and Publisher.
• Yahoo Scout shows AI search doesn't have to cannibalize publisher traffic. Playwire.
• Digital News Report 2026, episode 4, how people are using AI chatbots for news. Reuters Institute.

Dispatch
Several publishers are choosing engagement over exclusion. Reuters launched a Model Context Protocol server that lets customers' AI agents access licensed Reuters content programmatically, turning machine access into a paid, controlled product rather than an open crawl. Playwire reported that Yahoo's Scout demonstrates AI search can send traffic back to publishers, and the Reuters Institute's Digital News Report examined how audiences actually use AI chatbots for news.

For strategists these items describe the constructive alternative to closing the gate. Reuters is selling machine access on defined terms, with authentication and subscription rather than free scraping, which converts an AI reader into a customer. Yahoo Scout suggests that a referral-preserving design is technically possible, which matters because it shows the extractive pattern is a choice rather than a law of the medium.

The pattern is the same equilibrium seen in the licensing chapter, now expressed in architecture. A publisher can let machines in through a metered, paid door that returns value, and the MCP model shows what that door looks like in practice. Adaptation and enclosure are two responses to one problem, and the sector is testing both at once.

Reflections
• What commercial terms turn machine access to content from a cost into a durable revenue line?
• If referral-preserving AI search is technically possible, why would a dominant platform choose a design that removes traffic?

19.11 — The nuclear option, and the wall with two gates
Source
• Once Unimaginable, Publishers Are Preparing to Opt Out of Google Search. Adweek.
• Cloudflare's latest AI rankings expose the web's biggest free rider. Business Insider.

Dispatch
Adweek reported that publishers are preparing what it calls the nuclear option, removing themselves from Google Search by blocking its crawler, because Google uses a single crawler to both index sites and train its AI models and will not license content.

From September 15, Cloudflare will set new sign-ups and free-tier customers to block multi-purpose crawlers on any page carrying ads, and USA Today Inc. said it is prepared to delist from Google within six to twelve months, having built licensing deals with Meta, Microsoft and Amazon instead. Business Insider, citing Cloudflare's crawl-to-refer data, reported that Anthropic's bots crawled roughly 2,800 pages for every referral in early July, improved from about 8,800 to one in April, while DuckDuckGo returned close to one referral for every three crawls.

For decision-makers this is the moment the bargain is renegotiated at the level of infrastructure. Stephanie Cohen of Cloudflare framed the aim precisely, a technical solution that lets a publisher remain discoverable without giving content away for free, which reframes crawling as a permissioned transaction rather than an open default. The crawl-to-refer ratios give publishers a number to point to, since a reader who never returns is a cost, and a business cannot subsidise a counterparty that takes thousands of times more than it gives.

The pattern completes the week. When the reciprocal exchange fails, the people who fund information build a boundary, and that boundary separates two kinds of visitor. Human readers remain welcome, and machine readers must identify themselves, accept rate limits and quotas, request permission and, where appropriate, pay. This is enclosure as a rational response to a broken bargain, and it is the direct answer to the measured traffic loss in the previous signals.
Reflections
• When a crawler returns one reader for every few thousand pages taken, at what ratio does open access stop being rational?
• How should a publisher weigh the reach of remaining in a dominant index against the revenue lost to zero-click answers?

Conclusion
The signals this week describe one structural shift. The open web was built on reciprocity, machines could read freely because they returned readers, and that reciprocity is now measured, priced and, in places, withdrawn. The randomised evidence on AI Overviews, the crawl-to-refer ratios from Cloudflare, and the licensing settlements from Getty to Reuters all point to the same recognition, that information is scarce and costly and must be paid for if it is to keep being produced.

The central question running through the week is who pays for knowledge once machines can read it without a reader ever arriving. Europe is answering through law, with Denmark and France insisting that content on a platform generates a payable claim, and the answer is spreading to Nigeria and to collective bodies such as SPUR. Publishers are answering through architecture, with Reuters selling metered machine access and others preparing to close the crawler entirely. Litigation, licensing and innovation are moving the sector towards an equilibrium in which access to quality journalism is a transaction rather than a gift.

That equilibrium produces a distinct image. The web becomes a walled library, welcoming to people and guarded against machines that will not pay. It has two gates. One is marked for human access and stands open, because readers are the point of journalism and always were. The other is marked as a machine gate, an API gate, where authorisation is required and rate limits, quotas and permissions apply, because a machine that consumes without returning value is a cost the commons cannot carry.

This wall is a warning rather than an accusation. It shows what follows when the largest gateways decline to share the value they capture, and it invites the firms that operate at that scale to set a constructive example by paying for what they use. Quality journalism is scarce and expensive, and it deserves proper remuneration, and the combination of claims, licences and new products can still produce a sustainable settlement in which both gates stay open on fair terms.

The illustration for this dispatch makes the settlement visible, a great wall drawn around the library of human knowledge, with a lit human gate that welcomes readers and a separate machine gate where access is authorised, metered and paid, a modern Chinese wall that keeps the commons open to people while asking every machine to identify itself, respect the limits and honour the value it takes.

(Synthetic Image)

This week the material gathers around this question: how does verified, human-made journalism keep its economic footing while generative AI reshapes discovery, distribution and the very inputs that models learn from. The links to the news stories are about the integrity of the information supply, the newsroom practices that adapt to AI, the economics of search and content payment, and finally the legal and regulatory architecture that will decide who is paid and on what terms.

Three chapters run through this dispatch.

The first concerns the trustworthiness of the sources that both people and machines rely on, from a reported Russian effort to poison reference material to the workshops where agencies build provenance standards for images.

The second concerns value: where audiences turn first, how much traffic search now retains, and what it costs to keep producing high-quality content at all.

The third concerns law: the lawsuits, licences and copyright reforms that together are searching for a workable price for journalistic work.

These themes connect directly to the strategic world that this dispatch serves. AI raises the productivity of newsrooms and lowers the friction of misuse at the same time. Platforms remain the largest gateway to audiences and carry a corresponding social responsibility. Regulators in Europe, Korea, India and the United States are each testing how existing frameworks apply to a technology that consumes content at industrial scale.

The dispatch closes with the EANA, the alliance of European news agencies, speaking with one voice, through a formal EANA response to the European Commission on the review of the Digital Single Market Directive.

The earlier chapters build towards that statement, because it frames the same tension in the language of policy: verified journalism is scarce and costly, and its continued supply depends on enforceable copyright and fair remuneration.

Chapter 1 — Information Integrity and Adversarial Risk

18.1 — Poisoning the sources that machines read

Source
• Russia launched secret “Project 2026” to make AI spread false information — RBC-Ukraine
• Crowdsourced validation is not fact-checking: why the distinction matters — EDMO

Dispatch
RBC-Ukraine, drawing on classified documents reported via Bloomberg, describes a Russian programme it calls “Project 2026,” run by an entity styled the Agency for Social Design, that aims to seed the reference layer of the internet with controlled material. The reported method is to build Wikipedia-like reference sites, fabricated think tanks and cooperative outlets so that the primary sources consulted by chatbots and search engines return distorted results. EDMO, writing separately, sets out why community validation of the Community Notes type differs from evidence-based professional fact-checking, and treats Meta’s 2025 shift towards crowdsourcing as a defining moment for the field.

For news organisations and policymakers the significance is important. If adversaries can shape the material that models treat as authoritative, then verification moves upstream, to the provenance and independence of sources, rather than to the individual claim. The value of an agency that maintains verified, attributable records grows in proportion to the volume of synthetic and manipulated material in circulation.

The broader pattern joins supply-side manipulation to the quality of machine outputs. A model is only as reliable as the corpus it consumes, and the corpus is now a contested space. Professional fact-checking and crowdsourced validation each have a role, and clarity about what each can and cannot establish becomes part of the infrastructure of trust.

Reflections
• What distinguishes professional fact-checking from crowdsourced validation in a way that a policymaker, an editor and an AI developer would all accept?
• How far should an editor rely on model outputs when the underlying corpus is known to be a target of manipulation?

18.2 — Standards and tools for verified content

Source
• News organisations gather in Paris to collaborate on securing media provenance — AFP
• How to Use AI to Help Find Civilian Harm — Bellingcat
• Two coordinated networks, one domestic, one foreign, target the same Philippine Facebook pages — DFRLab

Dispatch
AFP reports a Paris workshop convened with the BBC and Media Cluster Norway, supported by the IPTC, bringing together more than forty participants from over twenty organisations, among them CBC, Deutsche Welle, France Télévisions, ITV, NHK and Al Jazeera, to advance image provenance through C2PA and Content Credentials. Bellingcat describes a machine-learning approach, with a gradient-boosting model performing best, that ranks Telegram posts by the likelihood of documenting civilian harm across more than two thousand five hundred incidents in Ukraine, freeing researchers to concentrate on verification. DFRLab documents two coordinated networks, one domestic and one linked to Chinese Spamouflage activity, that targeted the same Philippine Facebook pages and both relied on AI-generated images.

The strategic implication is that AI serves both the defence and the offence of the information ecosystem. Provenance metadata gives audiences and machines a way to check where an image originated, while the same generative tools lower the cost of coordinated manipulation. Agencies that adopt Content Credentials and ranking tools convert a defensive obligation into an operational advantage.

The pattern places verification at the centre of news value. Provenance standards, adversarial detection and machine-assisted triage together form a practical response to synthetic media, and the organisations that help set those standards shape the terms on which trust is established.

Reflections
• What incentives would persuade platforms and AI developers to honour provenance metadata rather than strip it?
• Which standards deserve collective investment when adversaries adapt as quickly as detection methods improve?

Chapter 2 — Newsroom Practice and the Productive Use of AI

18.3 — Reclaiming time, keeping the rules simple

Source
• Reclaiming time for journalism — International Media Support (IMS)
• Sipa Ouest-France’s AI lessons: keep the rules simple and trust your people — WAN-IFRA
• Most newsrooms don’t have an AI problem, they have a coordination problem — JournalismAI

Dispatch
IMS profiles the Polish local outlet JAW.pl, which uses AI to scan sources and around two hundred leads a day, allowing journalists to spend their time on verification and fieldwork; the tools support the work rather than replace it. WAN-IFRA reports the experience of Sipa at Ouest-France, where about eighty per cent of staff already used AI before any formal policy existed, and where management settled on one clear rule, that editorial content stays within an in-house model called Muse, and involved unions early. JournalismAI, writing for the Strategy Lab, argues that individual AI use is now widespread while shared guidelines remain absent, so the pressing issue is coordination rather than capability.

For decision-makers this describes a governance choice made from a position of strength. Simple, enforceable rules paired with an in-house model address the reality of everyday use more effectively than elaborate frameworks that staff route around. Early engagement with unions and clear internal boundaries turn informal experimentation into a managed capability.

The wider pattern is that adoption precedes policy, and institutions that acknowledge this design for it. Newsrooms that build their own tooling and set a small number of firm rules choose to keep editorial value and control inside the organisation.

Reflections
• How should a news organisation decide which tasks to keep inside an in-house model and which to entrust to external providers?
• When informal AI use already exists across the newsroom, what does effective coordination require of editorial leadership?

18.4 — Building AI products across the profession

Source
• How Time and others are rebuilding parts of the web for AI agents — Digiday
• ABC News bringing more regional stories to audiences using AI — ABC (Australia)
• AP at Cannes: CEO Daisy Veerasingham stresses the essential role of facts in the age of AI — Associated Press

Dispatch
Digiday reports that Time is converting pages to markdown through TollBit and whitelisting selected bots, while The Economist and Le Monde experiment with WebMCP approaches to serve AI agents; the consultant Scott Messer cautions that building for agents becomes a pure cost where it produces no click and no return. ABC in Australia describes “ABC Assist,” which reformats verified radio journalism into digital articles with journalist approval, reporting time savings of up to half in a pilot across Western Australia and Tasmania. At Cannes, AP chief executive Daisy Veerasingham argues that trusted facts are essential in the AI age and notes that AP now serves the market of AI platforms alongside its traditional clients, describing a collective responsibility for the factual ecosystem.

The implication is that news organisations are building AI products as a strategic choice, to extend reach, to serve new machine clients and to raise internal productivity. Each institution is weighing where value accrues, whether from agent traffic that may not convert, from faster regional coverage, or from licensing verified facts to the platforms themselves.

The pattern shows the profession investing in AI on its own terms. Public broadcasters, publishers and agencies each build capability that reflects their mission and their audience, and the common thread is the priority placed on verified facts as the durable asset.

Reflections
• How should a publisher value agent traffic that arrives without a click and without a direct subscription relationship?
• When an agency sells verified facts to machines, how does it price the difference between raw data and edited, attributable journalism?

18.5 — AI as an instrument of investigation

Source
• A local newsroom traces the wounds Argentina’s dictatorship left on its community — LatAm Journalism Review
• An exposé by students highlights failings of mainstream journalism in India — Nieman Reports

Dispatch
LatAm Journalism Review describes how Qué digital, a local outlet in Mar del Plata, used Google tools including Pinpoint, NotebookLM and AI Studio to produce a documentary project on the wounds left by Argentina’s dictatorship, drawing more than twenty-five thousand views in its first week. Nieman Reports recounts how three seventeen-year-old students exposed flaws in the marking of India’s CBSE examinations, including a password set to “123456” and contract manipulation, while mainstream outlets downplayed the story.

For strategists these cases show AI lowering the cost of ambitious investigative work for small and local newsrooms, and they show the reputational stakes when established media miss a story that others surface. Capability and credibility move together, and the tools are available well beyond the largest institutions.

The broader pattern connects AI-assisted investigation to democratic accountability. When a local outlet or a group of students can assemble evidence at scale, the advantage held by well-resourced newsrooms narrows, and the premium shifts towards editorial judgement and the willingness to publish.

Reflections
• What can established media learn from the spread of AI investigation tools to small newsrooms?
• What editorial standards should govern AI-assisted documentary work so that method strengthens rather than weakens credibility?

Chapter 3 — The Economics of Discovery and Content

18.6 — Search, clicks and the cost of visibility

Source
• Info Ouest-France: Google’s AI Overviews feature will launch in France this summer 2026 — Ouest-France
• Google AI Overviews study finds lost clicks weren’t lower quality — Search Engine Journal
• Where do Americans turn first for information about breaking news — Pew Research Center

Dispatch
Ouest-France reports, that Google will launch AI Overviews and AI Mode in France by 23 September 2026, and that Google has written to some four hundred and fifty publishers promising control, transparency and remuneration under neighbouring rights; the article records that the New York Times chairman A.G. Sulzberger described AI practices as “a shameless theft” at a Marseille congress, while the Google France director offered a rebuttal. A study reported by Search Engine Journal, finds a 39.8 per cent drop in organic clicks where AI Overviews appear, with such summaries triggered on roughly 41 per cent of queries, and concludes that the lost clicks were of the same quality as retained clicks, which goes against Google’s claim that the affected clicks were low-value. Pew reports that 36 per cent of Americans turn first to a news organisation for breaking news, down from 54 per cent in 2018, with 28 per cent going to search, 19 per cent to social media and 1 per cent to AI chatbots, and with the under-thirties favouring social platforms.

For news organisations the economic stake is direct. AI summaries retain a larger share of attention within the search page, and the demonstration that the lost clicks carried real value strengthens the case for remuneration rather than for treating the change as a natural evolution of behaviour. The scale of Google as a gateway, and its stated commitments on transparency and neighbouring rights, place a responsibility on the largest platform to set a constructive example.

The pattern ties discovery to sustainability. Where audiences turn first determines the revenue that supports reporting, and the shift towards search summaries and social platforms concentrates that decision in a small number of intermediaries. Value created by journalism is being consumed at the point of discovery, which is precisely where the remuneration question arises.

Reflections
• How should a publisher weigh the promises of control and transparency from a dominant search platform against the measured loss of traffic?
• As younger audiences turn first to social platforms, what does a news organisation need to build to keep a direct relationship with them?

18.7 — Paying for content and pricing the reader

Source
• Cloudflare ties AI payouts to citations as 50% of crawls waste — PPC Land
• Cloudflare’s new policy pushes AI companies to pay for publishers’ content — TechCrunch
• Surveillance pricing laws are coming for dynamic subscription strategies — Digiday

Dispatch

TechCrunch reports that from 15 September 2026 Cloudflare will by default block mixed-use crawlers on pages that carry advertising, and will move from a “Pay Per Crawl” to a “Pay Per Use” model as part of what it calls Content Independence Day. PPC Land adds that Cloudflare is tying AI payouts to citations, with pay-per-query arrangements involving companies such as Ceramic.ai and You.com, and reports that more than half of well-behaved bot crawls re-fetch pages that have not changed. Digiday, in a briefing by Sara Guaglione, describes surveillance-pricing laws coming for subscriptions, including a New York “One Fair Price Act” that would bar pricing based on personal data, a Washington Post class action over renewals rising from sixty to as much as a hundred and seventy dollars, and precedents in Maryland and Connecticut.

For strategists these developments describe two edges of the same economic question: how content is paid for on the way in and how readers are charged on the way out. Yet the inbound side rests on a category error. Cloudflare is building a market for machine access to content on the assumption that the cost of that content is variable, that it can be metered, consumed and settled per use, per query or per crawl. It is not. The cost of journalism is overwhelmingly fixed. A newsroom, its editorial capacity, its verification infrastructure and its institutional standards must be paid for whether a given article is fetched once, a million times or never. A pricing model that ties revenue to variable usage is structurally incapable of recovering a fixed cost base. From twenty-five years in business intelligence and media monitoring, this is not a matter of opinion: pay-per-use and pay-per-query models cannot fund the fixed economics of newsgathering, and never have.

The second flaw is one of responsibility and control. A model that charges only for what is “actually used” places the metering, and the moral weight of selection, on the wrong party. No upstream provider can verify what a downstream system truly consumed once content has crossed the API boundary; the claim of usage-based fairness is therefore unauditable by design, and a business model that cannot be verified by the party being paid is, in practice, a corrupt one. The economics are simpler than Cloudflare’s machinery suggests. When you come to collect content through an API, you pay for it. The publisher is not responsible for the consumer’s selection, and the consumer’s added value, whatever they choose to build on top, is their responsibility to fund. Collection is the billable event. Whether the content is ultimately used or not, if you come to take it, you pay for it.

The pattern therefore places the value of journalism inside a maturing market, but only if the market prices the right thing. Payment for machine access and transparent reader pricing can move the sector towards treating content as a scarce and costly good rather than a free input, provided access itself, not speculative usage, is the unit of exchange. There is no such thing as a free lunch.

Reflections

• When the true cost of journalism is fixed, why does infrastructure-led metering keep reaching for variable, usage-based pricing, and who benefits from that mismatch?

• A pricing claim that the paying party cannot audit is not a fair market but a leap of faith. What verification standard should publishers demand before accepting any usage-based settlement?

18.8 — Regulating attention and the wider media economy

Source
• Getty Images scraps Shutterstock merger — Reuters
• Bookstores are staging a comeback — Axios
• Indonesia’s social media ban tests families’ digital reality — Nikkei Asia

Dispatch
Reuters reports that Getty Images abandoned its planned merger with Shutterstock, valued at some 3.7 billion dollars, after the United Kingdom Competition and Markets Authority required the sale of Shutterstock’s editorial arm over concerns about competition in news content. Axios, in a piece by Nathan Bomey, describes a turnaround at Barnes & Noble under James Daunt and a rise of about seventy per cent in United States bookstores between 2021 and 2025, noting that AI summaries are a headwind and citing Daunt’s view that AI will not replace books. Nikkei reports that Indonesia’s ban on social media for under-sixteens takes effect on 28 March 2026 under a regulation known as PP Tunas, covering high-risk platforms including YouTube, TikTok, Meta services, X and Roblox, and affecting some seventy million minors.

For decision-makers these items describe the regulatory and market forces that surround the attention economy. Competition authorities scrutinise concentration in news content, physical retail demonstrates continuing demand for edited, curated work, and a large state moves to restrict minors’ access to platforms, each of which shapes the environment in which journalism competes for attention.

The pattern links market structure to the sustainability of quality content. Where regulators protect competition in news, where audiences pay for curated books, and where states limit platform access, the premium on trusted, edited work holds firm even as discovery habits change.

Reflections
• If audiences continue to pay for curated books despite free AI summaries, what does that indicate about willingness to pay for edited journalism?
• How should news organisations plan for a generation of readers who reach adulthood under restricted access to the platforms that once drove discovery?

Chapter 4 — Copyright, Rights and the Search for Equilibrium

18.9 — Licensing, sovereign models and the price of consent

Source
• GPT-NL respects copyright, cui bono? Part 1 — Kluwer Copyright Blog
• GPT-NL respects copyright, cui bono? Part 2 — Kluwer Copyright Blog
• New copyright licence helps businesses manage AI content risk — AICD / Copyright Agency (Australia)
• No Data, No Party: why competition for high-quality journalism matters in the GenAI race — The Platform Law

Dispatch
The Kluwer Copyright Blog examines GPT-NL, the Dutch sovereign model developed with TNO on an opt-in, licensed-only basis, including a deal with ANP and NDP Nieuwsmedia, and observes that it underperforms the Swiss Apertus model while its revenue-share arrangements may amount to a share of very little; Kluwer argues for a deployment-based, non-waivable levy to replace the Article 4 opt-out, framing copyright as a right to be paid rather than a right to refuse.

The Australian Copyright Agency, reported through the AICD, has introduced a licence that lets businesses lawfully use Australian news content in generative AI prompts. The Platform Law, in “No Data, No Party,” treats high-quality journalism as a third essential input in the generative AI race alongside compute and talent, and frames access to it as a competition question.

For policymakers and news organisations these pieces set out the range of instruments now in play, sovereign models built on licensed data, collective licences for enterprise use, and competition analysis that recognises journalism as a scarce input. Each approach attempts to convert consent into a workable price, and each reveals the difficulty of doing so where the underlying revenue is uncertain.

The pattern shows licensing and legal design working towards equilibrium. Claims establish the boundaries, negotiations and licences establish workable prices, and from those two forces a sustainable business model can emerge, provided the levy or licence reflects the real value of the content rather than a token payment.

Reflections
• What licensing design turns the principle of consent into a price that sustains the production of high-quality journalism?
• When a sovereign model built on licensed data underperforms, while access to high-quality journalism is an essential input for AI, how should publishers and public funders weigh data sovereignty against economic sustainability?

18.10 — Courts, access and public trust

Source
• Publishers sue South Korea over the shift to AI digital textbooks — Chosun Biz
• “Methyl Isocyanate of Law”: Supreme Court quashes verdict based on AI-generated precedents — The Hindu
• “Go Pound Sand!”: local officials move to restrict journalists’ access — Columbia Journalism Review

Dispatch
More than ten publishers of AI digital textbooks are suing the South Korean government, seeking around 122 billion won, on the basis of legitimate expectations and the protection of trust after a policy reversal. The Hindu reports that India’s Supreme Court quashed a verdict that had relied on AI-generated fake precedents, describing the problem as the “Methyl Isocyanate of Law,” and notes a Bar Council committee and draft regulations on AI in courts for 2026. The Columbia Journalism Review, documents local United States officials restricting journalists’ access.

For strategists these cases show the legal system absorbing AI-related disputes across very different fields, from procurement and reliance on policy to the integrity of court records and the conditions for reporting. Each ruling and each restriction contributes to the framework within which journalism and AI will operate.

The pattern connects legal certainty to public trust. Courts that reject fabricated precedent, governments held to their commitments, and officials tested on access each define the reliability of the institutions that journalism reports on and depends upon.

Reflections
• What standard of verification should courts and newsrooms alike require before treating an AI-generated citation as reliable?
• How should news organisations respond when local officials narrow the access on which accountability reporting depends?

18.11 — Binding protection and the funding of the ecosystem

Source
• MPM2026 report calls on the EU to take binding measures to protect journalists’ working conditions — European Federation of Journalists / EFJ
• The democratic news ecosystem needs adequate funding, here’s how it can be done — EU DisinfoLab
• YulChon unifies AI-era compliance — Chosun Biz

Dispatch
The European Federation of Journalists reports that the 2026 Media Pluralism Monitor, produced by the CMPF, finds working conditions deteriorating in one country in three, with only six of thirty-one assessed countries rated as safe, and identifies AI as a pressure on staffing; the report cites Belgium’s Elle case, in which AI-generated content appeared under fictional journalists, and calls for journalists to be included in negotiations over platform and AI revenue.

EU DisinfoLab proposes an independent endowment for the news ecosystem, funded through the 2028 to 2034 Multiannual Financial Framework and an AgoraEU instrument on the order of 8.6 billion euros, modelled on the European Endowment for Democracy. Chosun Biz reports that the Korean law firm YulChon has created a dedicated centre to handle the overlapping regulation of AI, platforms and data.

For decision-makers these items describe the institutional response to AI pressure on journalism, through monitoring, proposed public funding and the growth of specialised legal capacity. Binding measures, an endowment and coordinated compliance each address a different part of the sustainability problem.

The pattern places journalism’s economic foundation within European policy design. Where monitors document deterioration, where funding instruments are proposed, and where legal expertise consolidates, the terms are being set for how a democratic information ecosystem will be financed and protected.

Reflections
• What binding measures would most improve journalists’ working conditions without displacing the editorial independence they are meant to protect?
• When AI-generated content appears under fictional bylines, what disclosure standard restores the trust that a byline is meant to carry?

18.12 — A common European position: the EANA statement

Source
• Protecting the European Information Ecosystem: Why Copyright Enforcement in the AI Era is Essential for Democratic Resilience and Innovation — European Alliance of News Agencies (EANA), in response to the European Commission’s Call for Evidence on the review of the Digital Single Market (DSM) Directive; contributors Alexander Giboi (EANA), Lilian Hübner (dpa), Verena Krawarik (APA)

Dispatch
The European Alliance of News Agencies, which brings together the leading agencies across the continent, has submitted a formal statement in response to the European Commission’s Call for Evidence on the review of the DSM Directive, welcoming the Commission’s commitment under Executive Vice-President Henna Virkkunen to foster technological sovereignty while protecting Europe’s democratic fabric. The statement describes news agencies as the bedrock of verified, fact-based journalism and as the first line of defence against disinformation, and warns that the rapid proliferation of generative AI models threatens the economic viability of this ecosystem. EANA calls for the DSM Directive to be adapted to enforce transparency, mandatory licensing and robust protection against the unauthorised exploitation of journalistic content, on the central principle that there should be no content without consent.

For news organisations and regulators the statement gives the week’s themes a coordinated European voice. It links the integrity of the information supply, the economics of discovery and the design of copyright law to one conclusion, that verified journalism requires enforceable rights and fair remuneration to remain viable at scale.

The statement rests the case on value creation. Verified, independent, high-quality news requires significant and continuing investment, and generative AI models rely on that premium content to train their systems and to keep their outputs factually accurate; a durable settlement therefore pays for the content it consumes rather than treating it as a free input. The map of EANA’s member agencies across Europe illustrates the scope of the ecosystem that the statement seeks to sustain.

Reflections
• How does the principle of no content without consent hase to be translated into enforceable obligations within a revised DSM Directive?
• When AI outputs depend on premium journalistic content for their factual accuracy, what remuneration model reflects the real value that content provides?

Conclusion

This week’s material describes some important insight: verified journalism is scarce and costly to produce, and its continued supply depends on being properly paid for. Adversaries work to corrupt the sources that people and machines consult, newsrooms adopt AI to raise the productivity of verified work, search summaries retain the attention that once funded reporting, and courts and regulators test how existing rights apply to a technology that consumes content at industrial scale.

Across these chapters the same three forces recur. Value is created by the human effort of verification and editing. Platforms remain the largest gateways to audiences and carry a corresponding responsibility to set a constructive example. Law provides the framework within which claims establish boundaries, licences establish workable prices, and a sustainable business model can emerge from the combination.

The European agencies, speaking through EANA, place that combination at the centre of their response to the Commission. Their principle of no content without consent expresses in policy terms what the economics already imply, that content used by AI systems should be licensed and remunerated rather than taken as a free good.

For news organisations, regulators, democratic institutions and the readers who care about the information they rely on, the practical task is to move litigation, licensing and innovation towards a shared equilibrium. That equilibrium recognises the value that verified journalism creates, pays for it accordingly, and thereby keeps the ecosystem that the EANA map depicts both independent and financially sound.

EANA

Four themes organise this edition.

The first concerns authorship and the duty to disclose.

The second concerns the legal aspects, where more than a hundred copyright actions, new European rules, and a fresh legal argument about how training data was obtained are bringing more equilibrium.

The third concerns innovation, where publishers and agencies build AI products on their own archives, their own sources and their own editorial standards.

The fourth concerns the economics underneath all of it, where the model of the referral click that financed the open web and its advertising model, is about to retire.

These themes meet at the gateway between readers and the journalism that informs them, and the engine that powers most of the old internet, Google, occupies much of that gateway.

The screenshot that closes this dispatch illustrates the case plainly: a reader asks what AI summaries are, and the AI overview explains them while the cited material remains almost invisible, with the prominent links pointing back into Google's own properties rather than to the creators who wrote about the subject. It is a useful warning, not an accusation, and it deserves to be read as one. It's never wise to bite the hand that feeds you.

So let's ask a few practical questions to news organisations, regulators and informed readers alike. In an information economy where attention, data and capital concentrate around a small number of platforms, how is the value of journalism measured, and through which mechanism is it priced? Let us follow the signal from the question of authorship to the question of remuneration.

Chapter 1 — Authorship and the Duty to Disclose

17.1 — When the byline no longer guarantees the author

Sources
Germany's media shaken by AI scandal, DW
Editorial: Are you a simulation? AI and the art of the interview, Chemical & Engineering News (American Chemical Society)
New York City Council Candidate Is Accused of Forgery Over AI-Generated Posts, U.S. News

Dispatch
Germany's press absorbed two breaches of editorial trust this week. Tagesspiegel suspended its former publisher after he admitted composing opinion pieces with AI and not declaring it, and the Frankfurter Allgemeine withdrew a guest op-ed that had been written with similar assistance. A C&EN editorial describes how the problem reaches the source as well as the author: verification specialists now suggest a simple physical test for whether an interviewee is a generated avatar, a practical illustration of how far the question of identity has shifted.
The common thread is the integrity of attribution as a managerial control. A former New York council candidate has been charged with forgery for fabricating endorsements and a counterfeit news article with AI, including a manipulated image of a rival.
For news organisations the operational lesson is that trust is located in the disclosure rather than in the tool. As the C&EN piece observes, asking a counterpart to confirm they are human may become an ordinary part of professional contact rather than a sign of suspicion.

Reflections
• If readers expect to verify the human behind a byline, what assurance should news organisations really give to earn trust?
• Who should carry the burden of proof, and how is that burden best organized?

17.2 — The composition of the source pool

Sources
Revealed: ChatGPT draws more on GB News, Al Jazeera and Marie Claire than the BBC, IPPR analysis shows, IPPR
The British government wants to force more trustworthy news into your doomscrolling, Nieman Lab

Dispatch
An IPPR analysis finds that ChatGPT draws on a narrow range of outlets, citing The Guardian in 58 per cent of answers while the BBC is absent because it has exercised its opt-out. The think tank argues that AI assistants now perform an editorial function through the sources they include and omit, producing a new distribution of visibility among news brands.

Days later the UK government published a Green Paper proposing to make public service media more prominent in social feeds and search, with legislation held in reserve should platforms decline to cooperate voluntarily.

For policymakers the two items describe one question from opposite directions. The selection of which journalism a citizen encounters is made by systems whose operators describe themselves as neutral intermediaries. The British proposal answers that concentration of selection power with a public criterion of prominence, which raises its own governance question about who defines trustworthiness.
The underlying issue is that editorial influence has migrated to actors who do not present themselves as editors. Whether the curator is a model or a ministry, the legitimacy of ranking news for citizens is the matter at stake.

Reflections
• How far can a democratic government specify which news is prominent before promotion shades into endorsement?
• When the composition of the source pool changes without announcement, how should the public become aware of what was left out?

17.3 — The industrial supply of synthetic deception

Sources
Fact-checkers seek to stem a global tide of online scams, Poynter
Cyber based influence campaigns 15th - 21st June 2026 Report, Cyfluence Research
ARIAM: Disney, BBC, and the New York Times Form AI Content Coalition, AI Films Studio

Dispatch
The production of synthetic deception has reached industrial scale. Poynter reports fact-checkers contending with AI-cloned celebrities promoting fraudulent investment schemes, while Cyfluence's weekly review records deepfaked politicians, manipulated activist footage and more than 2,800 AI-generated deepfakes documented around the US midterm contests despite new disclosure statutes. The same review cites the Reuters Institute finding that news trust has fallen to 37 per cent and that weekly use of AI chatbots for news has risen to 10 per cent.

In response, established institutions are organising. Disney, the BBC, the New York Times, the Financial Times and others have launched ARIAM, a cross-sector alliance to set standards on training-data consent, attribution and synthetic-replica rights. News media start to treat AI governance as their own responsibility and have begun to draft the rulebook collectively.

Verification still depends substantially on human judgement. A coalition of brands is a meaningful contribution to that imbalance, and it operates alongside, rather than in place of, regulation and platform design.

Reflections
• If fabrications can be produced faster than they can be examined, how should we organize the newsroom of the future and maintain our DNA of trustworthiness?
• What would persuade platforms whose revenue rises with engagement to slow the circulation of the content that generates it?

Chapter 2 — Law in Search of Equilibrium

17.4 — A hundred actions, and the search for balance

Sources
NYT CEO confident in legal battles against OpenAI/Microsoft, Perplexity, Axios
Coalition of hundreds of local and regional newspapers sues OpenAI and Microsoft, InsiderNJ
Anthropic faces another copyright lawsuit, more than 100 authors say it used pirated books, India Today
Parallel litigation in 100-plus copyright suits v. AI companies produces litigation quagmire, resolution looks far off, ChatGPTisEatingTheWorld

Dispatch
Copyright disputes around AI have multiplied into a sustained body of litigation. The New York Times told Axios it remains confident in its actions against OpenAI, Microsoft and Perplexity, having committed around 20 million dollars over two and a half years, and a coalition of nearly 400 local and regional newspapers filed a further claim alleging systematic copying of their reporting.

A public tracker now counts 118 copyright actions against AI companies in the United States, with OpenAI named in 24, Meta in 14 and Anthropic in 10, even as more than 100 authors decline the Anthropic settlement in order to bring a separate claim.
Litigation and licensing operate as two halves of the same process: claims establish the legal boundaries, negotiated agreements establish workable prices, and a durable business model emerges from the interaction of the two.
This is how a developing market finds its footing. A thesis pressed in the courtroom and an antithesis negotiated at the licensing table together move the sector toward a synthesis, which is a settled and sustainable relationship between those who produce journalism and those who use it.

Reflections
• If litigation and licensing each supply part of the answer, how should a publisher sequence the two to reach a fair equilibrium?
• When a settlement in one case prompts fresh claims in another, what would a genuine resolution for the sector look like?

17.5 — The legal question of how data was obtained

Sources
How Strike 3 v. Meta Reframes AI Copyright Liability, Rain Intelligence
Belgian company sues Nvidia for training AI on its music catalog without permission, The Brussels Times

Dispatch
In Strike 3 Holdings v. Meta Platforms, the judge declined to dismiss the complaint and allowed it to proceed to discovery. The plaintiffs allege that IP addresses associated with Meta used BitTorrent between 2018 and 2025 to download their films thousands of times and to redistribute copies to other users. The court held that the reproduction and distribution involved in such downloading and uploading constitute an infringement under the Copyright Act in their own right, independently of whether the works were subsequently incorporated into any specific training set.

This reasoning introduces a brand new legal path. Rather than turning solely on what a model learned, a claim may rest on whether the underlying material was obtained lawfully, which is a separate and arguably more concrete question of evidence.

The Belgian company Winamp has filed in the United States, after a parallel action before the Ghent Commercial Court, over some 55,000 tracks from its catalogue said to have been used without authorisation, alleging copyright and contract violations together with unjust enrichment.
The development directs attention to provenance as a matter of legal record. Where the lawfulness of acquisition can be examined directly, the origin of the data becomes as material to the case as the behaviour of the model trained on it.

Reflections
• What incentive does a clean licensing route create once the manner of obtaining content carries independent legal weight?
• When the evidence consists of patterns of access rather than passages of text, which expertise can a court call on to assess it?

17.6 — Europe sets the framework

Sources
European Parliament and House of Lords publish reports on AI and copyright: policy directions for AI and copyright in the EU and the UK, Journal of Intellectual Property Law & Practice (Oxford University Press)
The Corner Newsletter: Our Next Revolution Conference on Capitol Hill & Europe's New Rules to Support Publishers in AI Era, Open Markets Institute
Swiss Parliament considers adding AI apps to media copyright bill, SWI swissinfo.ch
UK news industry backs law to stop deceptive AI scraping, Press Gazette
Here's What You Need to Know About Ottawa's New Policies on Social Media and AI, ChrisD.ca

Dispatch
European institutions are moving toward a licensing-first settlement. A comparative study finds the European Parliament and the UK House of Lords converging on a shared approach: maintain the opt-out under the CDSM Directive, strengthen transparency obligations, and cultivate a functioning licensing market rather than a broad exception.

Concrete measures are arriving alongside the principle. Open Markets reports Italy enforcing publishers' right to negotiate compensation, upheld by the Court of Justice, while the UK Competition and Markets Authority requires Google to let publishers opt out of AI summaries and model fine-tuning without consequent downranking, and to provide clear source attribution.

Switzerland is now considering whether to bring AI applications within the scope of its related-right copyright bill, which would oblige large online services to remunerate media companies for the use of their content, and the UK news industry supports a private member's bill requiring bots to identify themselves and their purpose.

The dynamics here are political as much as legal. French publishers, through the SNE, argue that lawmakers weakened a bill establishing a presumption of use under pressure to position France as a centre for AI, while Canada is advancing a broad set of digital-safety, privacy and deepfake rules expected to take full effect by 2028. The shared intention across jurisdictions is to make the use of journalism by AI both visible and remunerated, even where the pace of legislation differs.

Europe is constructing a framework in which access to journalistic content carries a price and a record. The question this raises is an economic one: whether content that is costly to produce can continue to be produced if its use generates no return, and on what terms that return should be set.

Reflections
• When licensing becomes the statutory norm, how is economic value determined when one party controls distribution and the other holds the content?
• How to enable a publisher to create value while avoiding that a dominant player can render this value commercially impractical?

Chapter 3 — Innovation: Building on Editorial Foundations

17.7 — Publishers and agencies build their own AI products

Sources
AFP and Mistral AI announce global partnership to enhance AI responses with reliable news content, AFP
VRT explores AI-driven conversational news formats, VNewsGuard Launches First AI Chatbot Built to Deliver Trusted Journalism Only from Reliable News Websites, NewsGuard
AI and the future of journalism: Balancing innovation and responsibility, European Science-Media Hub

Dispatch
Alongside the legal activity, several institutions are also building. AFP signed a multi-year partnership giving Mistral's assistant Le Chat access to its 2,300 daily stories in six languages, which AFP frames as both a diversification of revenue and a commitment to a European, verified-source approach.

VRT is developing conversational news grounded entirely in its own archive, allowing audiences to ask follow-up questions on subjects such as Ukraine and to receive answers drawn from VRT NWS journalism.

NewsGuard launched an assistant that responds only from 12,000 vetted publishers and commits to sharing revenue equally with the outlets it cites.

For news organisations this is innovation in the constructive sense. If audiences increasingly receive answers through conversational interfaces, there is clear strategic value in ensuring those answers originate in one's own journalism, carry one's own name, and generate a return. The European Science-Media Hub frames the discipline that should accompany this, namely a sustained attention to which systems are used, on whose data, and under whose accountability.

The opportunity is to move from being a source of raw material to being a recognised supplier of curated, attributable content. The accompanying consideration is the design of balanced relationships, in which a supplier maintains optionality across several partners and standards rather than depending entirely on one.

Reflections
• If your journalism informs another organisation's agents, what contractual terms keep the relationship a partnership of equals?
• When "trusted source" becomes a product attribute, who is qualified to certify reliability, and how is that certification kept accountable?

17.8 — Verification as a shared capability

Sources
Fake-o-Meter: Using AI to turn Confusing Claims into Clear Conversations, DW Innovation
No Data, No Party: Why Competition for High-Quality Journalism Matters in the GenAI Race, The Platform Law Blog

Dispatch
DW Innovation, the "Fake-o-Meter", running until 2029, aims to give ordinary users a conversational companion that examines images, video and audio and explains the context around a claim at the user's own pace.

The Platform Law Blog sets out the structural argument behind such efforts: high-quality, human-generated journalism is a third input in generative AI, alongside compute and talent, and access to it is a matter of competition policy as well as copyright.

For strategists the two pieces connect a practical tool to an economic thesis. Reliable content contributes materially to the accuracy and usefulness of models, which gives it genuine competitive value. The relevant advantage here is qualitative: specialised, verified and current reporting is distinctive precisely because it is difficult to produce, and that scarcity is what gives it bargaining weight regardless of the sheer volume of data a larger party may hold.

The constructive reading is that quality content is an asset whose worth will be recognised in the design of both markets and regulation. A healthy ecosystem rewards the production of distinctive journalism, which serves the interests of the developers who rely on it as much as the publishers who create it.

Reflections
• If quality journalism is a competitive input, how should it's valorisation be treated in competition and AI regulation?
• When distinctive, specialised reporting is a scarce resource, how do publishers best capture that value?

Chapter 4 — The Business Case: Google and the Economics of the Click

17.9 — The economics of the answered query

Sources
Google's AI Overviews are killing your organic clicks, Bright Labs
AI bots are reshaping the internet as human traffic loses dominance, Business Standard
Google AI changes could deal further blow to publisher Discover traffic, Press Gazette
Google's VP of Search tells CMOs: good SEO is still all you need for AI, ppc.land

Dispatch
The figures describe a measurable change in how value moves across search. Bright Labs, analysing more than 100 million keywords, reports organic clicks down over 60 per cent where AI Overviews appear, so that ranking first now yields roughly the traffic that ranking third produced before.

Cloudflare's data show bots generating 57.5 per cent of web requests, the first time automated traffic has exceeded human traffic, and Google's crawl-to-referral ratio moving from about two pages fetched per visitor sent to between fourteen and eighteen.

Press Gazette adds that Google has begun grouping publishers beneath AI summaries within Discover, a feed that has supplied a significant share of referral traffic for many newsrooms.

Against this, Google's VP of Search advises CMOs that sound foundational SEO remains sufficient, since AI Mode and AI Overviews operate on the same ranking systems.

For publishers the more nuanced reading is that optimising for a results page that increasingly resolves the query within itself produces diminishing referral returns, even as that same page may serve readers well.
The economic shift is that content continues to circulate while the associated referral traffic is redistributed. The relationship between content supplied and audience returned is being reset, and that reset is the substance of the business question facing publishers.

Reflections
• If a platform now captures the traffic that the source once did, how should a publisher reevaluate the relationship with such a platform?
• How long can a strategy built on referral remain viable when the results page increasingly answers the query in place?

17.10 — Google revises the terms of licensing

Sources
Google strikes tough negotiating stance with publishers over AI licensing, The Information
Google Looks To Reportedly Broaden Publisher Licensing Deals, MediaPost
Google launches initiatives to help Malaysian newsrooms navigate AI, changing audience habits, Malay Mail

Dispatch
As the value of the referral changes, Google is revising the terms on which it licenses content. The Information reports a firm negotiating position, under which broader licences, including consent to use content for model training, are presented as the condition for participating in the News AI pilot, while publishers who decline are told they will lose the payments previously made under the Showcase programme, which Google intends to wind down.

MediaPost adds that these comprehensive agreements consolidate legal and operational advantages for Google, and quotes an executive who notes that low referral volumes have led some publishers to develop direct subscriber and registration relationships.

In Malaysia, Google has launched Project Sigma 2.0 and Project Berita to help newsrooms adapt to AI and changing audience habits through training and tooling.

For news leaders the structural feature is that one party occupies several roles at once: it shapes discovery, sets the licensing terms, administers the legacy payments, and funds the capacity-building programmes. That concentration of roles is the analytical point, and it warrants careful contractual attention rather than indignation. It's worth recognising that a platform of this scale carries a corresponding responsibility, and that the largest gateway is the one best placed to set a constructive example.

The pivot some publishers are making toward direct subscriber relationships is best read as a calculated strategic choice rather than a retreat. Building owned audiences, first-party data and recurring revenue is a basic economic response to a changing distribution environment and a source of resilience in its own right.

Reflections
• If one party shapes discovery, licensing and training terms together, how should a publisher structure the agreement to preserve the balance?
• When the largest gateway also offers to help newsrooms adapt, how is it ensured that "navigating AI" serves the publisher's interest as well as the platform's?

Conclusion

The question of authorship concerns who may speak in a newsroom's name; the question of law concerns who may use its work and on what terms; the question of innovation concerns whether publishers can supply AI systems rather than merely feed them; and the question of economics concerns how value is shared once readers receive their answers on the results page.

Each leads back to the gateway between audiences and journalism, where attention, data and capital concentrate around a small number of platforms. Google occupies much of that gateway, and with that position comes a responsibility including the expectation that it sets a constructive example.

The screenshot that anchors this dispatch illustrates the point. A reader asks Google what AI summaries are, and the AI overview supplies the explanation while the cited material remains invisible and the prominent links lead back into Google's own services.

It is a clear illustration of the question at the center of this new economy.

The constructive path runs through synthesis. A sustainable future for journalism and for AI alike will be based on the premise of an equilibrium.

Let's be straightforward: quality journalism is scarce, costly to produce and valuable to those who use it, and a durable market is one that prices it accordingly.

The central question of Dispatch 17 is not who wins, but how the parties reach an equilibrium that keeps quality journalism worth producing.

Copyright APA

In Vienna this spring, in the gilded rooms of the Hofburg, the future of Austrian public service media felt suddenly very concrete. At the reception hosted by President Alexander Van der Bellen, I had the privilege of congratulating Clemens Pig on his ambition to become
the future Director General of ORF, while shaking the President’s hand and realizing, very literally, how small our European media ecosystem actually is.
This is not just another personnel move in a national broadcaster. It is the moment when the operating system of a news agency meets the political and financial stress test of a public broadcaster under pressure.

A mandate that begins with subtraction
Clemens inherits an institution that will start his mandate with a large minus sign on its balance sheet. From 2027 onwards, the Austrian federal government intends to cut a yearly 93 million euro compensation related to the VAT treatment of the new ORF contribution. This is not an accounting footnote; it is a structural shock that will require deep re‑prioritisation, cost discipline and a strategic re‑design of ORF’s portfolio, from linear TV to digital services and regional offerings.
Every European public broadcaster knows the pattern. First, the political narrative of “efficiency” and “belt‑tightening”. Then, the slow erosion of the capacity to invest in investigative journalism, culture, regional presence and digital innovation. In Austria, the discussion is made even more acute by the simultaneous increase of media subsidies elsewhere in the system, while ORF is asked to do more with substantially less.
For an incoming CEO, this is the equivalent of taking command of a ship while the crew is told that one engine and several decks will be removed mid‑voyage.

From APA to ORF: the quiet power of the agency mindset
What makes Clemens’ move so interesting is precisely that he does not come from within the ORF hierarchy. He comes from the DNA of a cooperative news agency: APA, owned by ORF and the Austrian daily newspapers, where he has been the architect of a transformation that combined technology, shared services and an explicit commitment to public value journalism.
In a news agency, the business model forces you to think in systems, not shows. You optimise infrastructure, data flows, content pipelines and risk management across clients who are simultaneously your owners and your competitors. You learn to be radically boring where it matters (uptime, security, compliance) and quietly innovative where it creates leverage (automation, AI‑assisted workflows, data products). That mindset is exactly what a public broadcaster in fiscal and political turbulence now needs.
Clemens has argued consistently that European media should approach AI and platform power with cooperation rather than fragmented competition, stressing the need to pool data, technology and standards instead of fighting alone against structurally dominant US platforms. That is pure agency logic: no single player is big enough, but networks can be.

ORF as “platform of society” – if the budget allows
In his public statements after the vote in the ORF Stiftungsrat, Clemens sketched an ambition to turn ORF from a traditional broadcaster into a “platform of society”, with a structure that puts Finance, Technology & Innovation, Program & Brands, and Audience & Platforms on equal strategic footing. This is more than an org chart; it is an attempt to encode in governance the reality that audiences are now distributed, streaming is default, and public value is negotiated every day, not only at 8pm in the main news bulletin.
The problem, of course, is that platforms are expensive – especially when they are supposed to serve the entire population, safeguard democratic discourse, protect minors, comply with a growing stack of EU regulation (DSA, DMA, AVMSD, GDPR, AI Act), and invest in cybersecurity and resilience by design. A 93 million euro annual cut is, in that light, not a “saving”, but a forced re‑scoping of what “platform of society” can realistically mean.
Here, the APA experience is again relevant. Under Clemens’ leadership, APA invested early in data services, media monitoring and technological infrastructure that could be shared across clients, including ORF itself. The same logic could help ORF build a modular, interoperable platform architecture that reduces duplication, increases resilience and allows scarce innovation resources to be focused where they matter most.

Cybersecurity, trust and the new attack surface of public media
Readers of this dispatch know that I usually approach media stories through the lens of cybersecurity and systemic risk. ORF is not just another broadcaster; it is critical democratic infrastructure. When you add budget cuts, political pressure and accelerated digitalisation, you also enlarge the attack surface – technical, informational and institutional.
On the technical side, a more data‑driven, streaming‑centric ORF implies complex identity management, expanded content delivery networks, and more integration with third‑party platforms. That requires a security architecture that treats every integration as a potential vulnerability and every newsroom workflow as a potential vector for manipulation or data exfiltration.
On the informational side, the core asset to protect is not a server, but trust. Der Standard rightly points to the polarised debates around ORF, the politicisation of governance, and the expectations that the new General Director must rebuild and defend the broadcaster’s credibility in an environment of social media outrage cycles and disinformation campaigns. Trust, in this context, is a security perimeter.
And institutionally, the combination of external budget decisions and internal restructuring will create uncertainty among staff and leadership layers. That is when phishing campaigns, social engineering and insider risk thrive. A news‑agency‑trained CEO, used to thinking in terms of operational continuity and redundancy, has a better starting point to treat resilience as a core function, not as a side project.

What travels from APA to ORF – and what cannot
So what exactly can travel from the APA playbook into the ORF environment?
• A network mindset: designing ORF as a node in a European information and technology network, not as an isolated national fortress.
• A data and infrastructure orientation: seeing content, archives, and metadata as strategic assets that can be shared, licensed and recombined under strict governance, instead of being locked in siloed systems.
• A culture of cooperative competition: balancing the role of ORF as shareholder, client and competitor in relation to other Austrian media, and translating that into transparent governance, fair access and non‑discriminatory data‑sharing frameworks.
• A pragmatic approach to AI: using automation and machine learning to enhance journalistic workflows, reduce repetitive tasks and improve monitoring, while defending human editorial responsibility and explainability standards that are aligned with emerging EU regulation.
What cannot simply be copied is the relative political insulation that a news agency still enjoys compared to a public broadcaster whose leadership appointments are themselves objects of party competition. ORF is broadcast politics, not just media infrastructure. Here, Clemens’ own insistence that ORF must remain “unabhängig von Parteien, Interessen und einzelnen Milieus” will be tested not in strategy slides, but in the daily grinding reality of appointments, coverage decisions and budget trade‑offs.

A handshake in the Hofburg
At the end of the reception with President Van der Bellen, there was a brief, symbolic moment: a handshake, a few words about Europe’s media landscape, and Clemens standing a step behind us, already carrying the weight of an institution that still defines how millions of Austrians see their country and the world. The photograph below captures that moment as a reminder that democratic media are ultimately built on human relationships, shared standards and a willingness to carry responsibility in public.
In cybersecurity, we often say that “attacks follow value”. In European democracies, public service media remain high‑value targets: for budgetary re‑allocation, for cultural wars, for disinformation, and increasingly for technical attacks. The question is whether we can turn this moment of vulnerability into an opportunity for institutional hardening, regulatory clarity and renewed public engagement.
As Clemens prepares to move from the agency universe to the public broadcaster’s bridge, I can only wish him all the courage, clarity and resilience this moment requires. Having watched him build APA into a modern, cooperative backbone for Austrian and European media, I am convinced he is the right person in the right place at exactly the right time – and I look forward to seeing what he will make possible with and for ORF.

Sources
• Heute.at – Budget-Hammer: ORF muss auf 93 Millionen verzichten

• MeinBezirk – Budgetsanierung: ORF muss massiv einsparen, Medienförderung steigt

• Kleine Zeitung – Bundesregierung will 93 Mio. Euro beim ORF einsparen

• MeinBezirk – Tiroler Clemens Pig ist neuer ORF-Generaldirektor

• Deutschlandfunk – Medienmanager Clemens Pig zum neuen ORF‑Generaldirektor gewählt

• Heute.at – Neuer ORF-Chef: Clemens Pig hat nach Stunden Mehrheit

• Joyn News – Clemens Pig zum neuen ORF‑Generaldirektor gewählt

• diemedien. – Clemens Pig (ORF)

• Kurier – Der neue ORF-General: Wer ist Clemens Pig?

• Der Standard – Regierung will 93 Millionen Euro beim ORF einsparen

• Der Standard – Was Clemens Pig mit dem ORF vorhat

Tomorrow, I will join ENT.A and the Lab of Tomorrow, part of Tomorrowland for a conversation with people who live from creativity, technology, performance, experience and intellectual property. That setting matters. ENT.A is part of a new Flemish effort to connect media, entertainment, gaming, culture, technology and knowledge institutions around innovation in the experience economy. Lab of Tomorrow, rooted in the Tomorrowland ecosystem, is built around the future of entertainment tech. So this is not just another debate about AI in news. It is a debate about what happens when every creative industry discovers that its work can be copied, recombined, summarised, distributed and monetised by machines.

That is why this dispatch leaves the broader geopolitical map aside and stays inside one pressure zone: journalism and the wider creative economy. The question behind tomorrow’s conversation is simple, but brutal: when machines can generate, summarise and distribute content at near-zero cost, what does a creator, a publisher, a journalist or a news agency still own, and what can they still charge for?

The central tension this week is the gap between value created and value captured. Audiences are drifting from news sites to social feeds, video and chatbots; trust is falling; search is turning into an answer engine; and AI systems are becoming increasingly hungry for verified, structured and reliable content. The people and organisations doing the original work are losing readers, clicks, leverage and sometimes jobs, while the platforms and systems that summarise them keep the attention and the advertising.

Five themes run through the links. First, the collapsing demand side, where audiences migrate away from the institutions built to serve them. Second, the fight over the pipes, where copyright, contracts and licensing are being rebuilt around AI access. Third, the answer engine, where search stops sending traffic and starts keeping it. Fourth, the machine inside the newsroom, where agentic AI promises efficiency but also redistributes power. Fifth, the flood of synthetic content, where provenance and trust become the real scarce goods.

For a press agency like Belga, sitting upstream of publishers and broadcasters as a piece of critical information infrastructure, the stakes are sharper still. Agencies sell exactly what machines most want to ingest for free: verified, structured, reliable fact.

So this dispatch is less a survey than a reality check. The creative industries are about to learn what newsrooms already know: the hardest part of the AI transition is not building the tool. It is deciding who gets paid when the tool works.

Chapter I
On the Vanishing Audience

16.1 — The Audience Walks Out

Source
The full title of the main article is: Publishing trends for 2026: Tech platforms overtake publishers as global news source, Press Gazette, by Charlotte Tobitt
https://pressgazette.co.uk/publishers/digital-journalism/news-publishing-trends-for-2026/

News sites are the new newspapers: People are abandoning them for social media, Nieman Lab
https://www.niemanlab.org/2026/06/news-sites-are-the-new-newspapers-people-are-abandoning-them-for-social-media/

Reuters Digital News Report 2026: Key takeaways for newsrooms and journalists’ unions, International Federation of Journalists
https://www.ifj.org/es/sala-de-prensa/noticias/detalle/category/ai/article/reuters-digital-news-report-2026-key-takeaways-for-newsrooms-and-journalists-unions

Dispatch

The 2026 Reuters Institute Digital News Report, read across these three accounts, marks a threshold: social and video platforms are now used for news by 54 percent of people across 48 markets, edging past television at 52 percent and news websites at 51 percent, with TikTok up fifteen points in five years to 22 percent. Trust in news has fallen to a record low of 37 percent globally, while interest in news in the UK has dropped from 70 percent in 2015 to 37 percent today, and only 17 percent of people paid for online news in the last year. The same report that documents this flight also notes a quieter fact: only 1 percent treat chatbots as their primary news source, so the audience is not leaving for AI, it is leaving for other humans on other platforms.

For news organisations this is the demand-side equivalent of a slow bank run, because the institutions that fund original reporting are losing the direct relationship with the people who consume it, and intermediaries now stand between the newsroom and its public. For a press agency the lesson is one level deeper, since agencies do not chase clicks but supply the publishers who do, and when those publishers shrink, the wholesale market for verified fact shrinks with them. The strategic question stops being how to win the next platform and becomes how to remain indispensable when no single platform commands the room.

The story is no longer that newspapers are dying; it is that the open news website, the great hope of the last two decades, is now ageing into the same decline. When the audience disperses to a dozen feeds, the scarce asset is not reach but the reason anyone comes back.

Reflections

• If the audience now lives on platforms a newsroom does not control, what is the one relationship a publisher must still own outright before everything else becomes negotiable?

• When trust and interest fall together, which of the two is the more dangerous to lose first, and why?

16.2 — Loyalty Over Reach

Source
The full title of the main article is: News Media Audience Trends Show Nuanced Story, Mather Economics, by Luke Magerko and Peter Doucette
https://www.linkedin.com/posts/luke-magerko-3072a35_may-news-media-audience-benchmark-report-activity-7472302033753010178-HfPl

DNR: What audiences actually need from you right now, Journalism.co.uk, by Marcela Kunova
https://www.journalism.co.uk/dnr-what-audiences-actually-need-from-you-right-now/

Dispatch

Beneath the gloom sits a more nuanced picture. Mather Economics reports that in May unique visitors fell 12.5 percent year on year and pageviews fell 5.9 percent, yet engagement held steady or improved across most segments, because publishers are losing casual readers faster than loyal ones, and attention is concentrating around brands with strong habits and direct relationships. German and Canadian publishers, with higher visit frequency, are cited as proof that decline is a choice of model, not a law of nature.

This reframes the entire crisis for decision-makers, since it suggests the future belongs less to publishers with the most traffic and more to those with the most loyal audiences, which changes what a newsroom should measure, reward and defend. The practical reading from the Digital News Report points the same way: audiences still want clarity, usefulness and a sense that someone is paying attention to what they actually need, rather than to what the algorithm rewards.

The metric that mattered for twenty years, scale, is quietly being replaced by the metric that pays the rent, loyalty. Reach was rented from platforms; loyalty has to be earned, and it cannot be repossessed overnight.

Reflections

• If casual reach is leaving anyway, what would a newsroom build differently the morning it decided to optimise for loyalty instead of traffic?

• Which habits keep a reader returning when every competing feed is engineered to keep them scrolling elsewhere?

16.3 — A Tale of Two Belgiums

Source
The full title of the main article is: Belgium (Digital News Report 2026), Reuters Institute for the Study of Journalism, by Ike Picone
https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2026/belgium

Digital News Report Nederland: Zorgen om online mis- en desinformatie nemen sterk toe in 2026, Commissariaat voor de Media
https://www.cvdm.nl/nieuws/digital-news-report-nederland-zorgen-om-online-mis-en-desinformatie-nemen-sterk-toe-in-2026/

Dispatch

Close to home, the picture sharpens. In Belgium, overall trust in news has fallen to 39 percent, but the national average hides a widening fault line, with Flanders at 49 percent and Wallonia at 28 percent, while only 14 percent pay for online news and social media use for news has risen to 41 percent. Across the border, the Dutch report finds half the population now seriously worried about mis- and disinformation, over a million people relying on social media alone for news, and 46 percent of under-25s following independent news creators rather than established brands.

For a Belgian press agency this is not abstract, because the linguistic trust gap means the same verified report lands in two communities with very different predispositions to believe it, and a single national supply of fact now meets a fragmented national appetite for it. The Dutch anxiety about disinformation is, paradoxically, an opening, since a population that is worried about being misled is a population that may once again value institutions whose entire business is not being wrong.

Two neighbouring markets, one lesson: trust is now local, uneven and contested, and the agency that treats its country as a single audience may be selling to a public that no longer exists.

Reflections

• When trust in the same facts differs by twenty points between two language communities, what is the required strategy for national news institution actually in order to be serving the country?

• If anxiety about disinformation is rising, how can a press agency turn that fear into a reason to pay for verified fact rather than another reason to disengage?

Chapter II
On Owning the Pipes

16.4 — Putting a Price on the Crawl

Source
The full title of the main article is: Publishers to bill AI firms for unwanted scraping, and take them to court if they don’t pay, Press Gazette, by Dominic Ponsford
https://pressgazette.co.uk/news/publishers-to-bill-ai-firms-for-unwanted-scraping-and-take-them-to-court-if-they-dont-pay/

Publishers Take Control Over AI Harvesting with New Contract, Movement for an Open Web
https://movementforanopenweb.com/publishers-take-control-over-ai-harvesting-with-new-contract/

AWS WAF adds AI traffic monetization capability to help content owners charge AI bots for content access, AWS News Blog, by Esra Kayabali
https://aws.amazon.com/blogs/aws/aws-waf-adds-ai-traffic-monetization-capability-to-help-content-owners-charge-ai-bots-for-content-access/

Dispatch

A new toolkit for charging the machines arrived this week. Thirty-one UK publishers, backed by the Movement for an Open Web, have adopted Search-Only Contracts that price unauthorised AI scraping at roughly 500 pounds per article and route disputes through small-claims courts, while the Movement reports that bots now make up nearly half of all web traffic, 99 percent of it unwanted. In parallel, Amazon Web Services has turned the same problem into a feature, letting content owners charge AI bots per request at the network edge, settled in stablecoin, on the blunt premise that AI crawlers already exceed half of many sites’ traffic and are growing more than 300 percent a year.

The strategic shift here is from prohibition to pricing, because for two years publishers tried to block the bots and lost traffic doing it, and the new move is to let them in but make them pay, converting an act of theft into a line of revenue. This favours those who can negotiate or instrument at scale, and a press agency, whose entire output is the high-value structured data these systems crave, is unusually well placed to set a price rather than absorb a loss.

The wall did not work, so the industry is building a tollbooth. The question is no longer whether to let the machines through, but who collects the fare, and whether the smallest publishers ever see a cent of it.

Reflections

• If access becomes a metered toll rather than a locked door, who sets the price, and what stops the largest platforms from simply pricing the smallest publishers out of the negotiation entirely?

• What does a press agency gain, and what does it risk, by being the first in its market to put a transparent price on machine access to its wire?

16.5 — The TDM Equation

Source
The full title of the main article is: The TDM equation: Why a EUR 600 billion share of Europe’s AI opportunity hinges on text and data mining, Implement Consulting Group, by Martin H. Thelle, Nikolaj Tranholm-Mikkelsen and Mads Sigurd Franch Andersen
https://cms.implementconsultinggroup.com/media/uploads/articles/2026/The-TDM-Equation/The-TDM-Equation.pdf

Dispatch

A consulting study, commissioned into the heart of the European copyright debate, puts a number on the other side of the argument. It estimates generative AI as a 1.65 trillion euro annual opportunity for the EU and warns that restricting commercial text and data mining could put 600 billion of it at risk, recommending that Europe preserve current TDM exceptions, lean on scalable opt-outs like robots.txt, and favour voluntary commercial deals over mandatory licensing. In short, it argues that the cost of protecting rights-holders too tightly is a slower, weaker European AI.

For news and creative organisations this is the framing they must contest in public, because it positions their content as raw material whose restriction is a drag on a continent’s growth, rather than as labour whose use deserves payment. The choice between an opt-out regime and a licensing regime is not technical; it decides whether the default is that your work is free to mine unless you object, or priced unless you agree.

Every figure in this report is an argument about who should bear the cost of the AI transition. The creative industries are being quietly cast as the friction in someone else’s growth story, and naming that framing is half the battle.

Reflections

• If the public debate frames protected content as a brake on Europe’s AI growth, how do news and creative organisations re-tell the story so that their work reads as value rather than friction?

• Which default serves the creator better, opt-out or opt-in, and who benefits most from leaving that question unsettled?

16.6 — From Lawsuit to Infrastructure

Source
The full title of the main article is: Shakespeare v. Anthropic: 100+ Authors Sue AI Giant, National Law Review
https://natlawreview.com/press-releases/shakespeare-v-anthropic-100-authors-sue-ai-giant

Warner Music’s Sureel Deal Signals the Next Phase of the AI Fight, The Streaming Wars
https://www.thestreamingwars.tv/news/warner-musics-sureel-deal-signals-the-next-phase-of-the-ai-fight/

Dispatch

Two stories from the wider creative economy show the copyright fight maturing in opposite directions. More than a hundred authors have sued Anthropic, alleging it trained its models on millions of pirated books from sources like Library Genesis, seeking up to 150,000 dollars per work and pointedly noting that the unresolved liability hangs over the company’s reported IPO. Warner Music, meanwhile, has bought the AI-detection firm Sureel to build “AI DNA” profiles of its catalogue, moving from litigation to owning the measurement layer on the logic that licensing requires verification, verification requires measurement, and measurement requires technology.

For decision-makers the contrast is the whole strategy lesson, because suing is how you establish that a right was violated, but building the infrastructure of attribution is how you get paid every time it is used thereafter. The creative industry is splitting into those who fight the last war in court and those who quietly build the toll roads of the next one.

Litigation sets the precedent; infrastructure collects the rent. The companies that win the AI era will be the ones that own the meters, not just the lawsuits, and a press agency built on attribution and provenance already speaks that language.

Reflections

• If the real money is in measurement and attribution rather than in lawsuits, where should a news agency be investing today to own the meter rather than merely win the case?

• What can journalism learn from the music industry’s pivot from suing the machine to instrumenting it?

Chapter III
On the Answer Engine

16.7 — When Search Stops Sending Traffic

Source
The full title of the main article is: Google’s AI Overviews killed 58 per cent of publisher clicks. Now it is adding a ‘Further Exploration’ section to bring some back, The Next Web
https://thenextweb.com/news/google-ai-overviews-publisher-links-search-traffic

AI Overviews erode publisher traffic as New York moves to unmask crawlers, PPC Land Notes, by Luís Rijo

PPC Land Notes

AI Overviews erode publisher traffic as New York moves to unmask crawlers

The relationship between artificial intelligence systems and the open web grew more adversarial across June 14, 2026. Two independent publishers documented steep traffic losses traced directly to Google’s AI Overviews. A bill advancing through the New York State Assembly moved to force AI crawlers into the open. Fresh measurement data charted how automa…

Read more

3 months ago · 1 like · Luís Rijo

Google is putting ads inside AI Overviews. Here’s why financial marketers should pay attention, Financial Marketer, by Andrew Frith
https://financial-marketer.com/google-is-putting-ads-inside-ai-overviews-heres-why-financial-marketers-should-pay-attention/

Dispatch

The engine that built the modern web is changing what it does with a click. Studies cited by The Next Web link Google’s AI Overviews to a 58 percent fall in publisher click-through, up from roughly 34 percent a year earlier, prompting Google to bolt on a “Further Exploration” links section to send some readers back out; PPC Land documents individual publishers losing up to 70 percent of search traffic. At the same time Google is beginning to place ads inside the AI Overviews themselves, which now appear in around two-thirds of question-based searches, completing the shift from a search engine that routes attention elsewhere to an answer engine that keeps it.

For publishers this dismantles the implicit bargain of the last two decades, in which they gave Google their content and Google gave them traffic, because the traffic is now being summarised away and, increasingly, monetised in place. For a press agency the danger is being rendered invisible twice over, once when its wire feeds a publisher’s page and again when that page is compressed into an answer that names no one.

Search used to be a door; it is becoming a destination. When the front page of the internet stops handing visitors to the people who reported the news, the open web’s founding deal is quietly being rewritten without the publishers in the room.

Reflections

• If the engine that once sent the traffic now keeps it, what is left to optimise for, and what should a publisher stop spending to chase?

• When an answer names no source, how does original reporting prove it was ever there, and to whom does it complain?

16.8 — Whose Words Are They Anyway

Source
The full title of the main article is: Landmark German ruling declares Google’s AI Overviews are Google’s own words and makes it liable for false answers, The Decoder, by Matthias Bastian
https://the-decoder.com/landmark-german-ruling-declares-googles-ai-overviews-are-googles-own-words-and-makes-it-liable-for-false-answers/

UK regulator sets out conduct requirements for Google’s search services, Reuters, by Muvija M.
https://www.reuters.com/world/uk-regulator-sets-out-conduct-requirements-googles-search-services-2026-06-17/

Dispatch

The law is beginning to answer back. A Munich court has ruled that Google’s AI Overviews are Google’s own words, not a neutral list of results, making the company directly liable for false statements they generate, on the reasoning that Google alone controls the algorithm and the presentation. Days later, Britain’s Competition and Markets Authority imposed conduct requirements on Google search under its strategic-market-status powers, demanding fairer ranking and data portability, and following earlier rules letting publishers opt out of AI features.

The strategic significance is that liability is migrating toward the summariser, because the moment an AI answer is treated as authored content rather than as a passive index, the protections that shielded platforms for two decades start to fall away, and so do the economics of answering for free. For news organisations this is leverage, since a system that must stand behind the accuracy of its answers has a powerful new reason to license the verified sources that keep it from being wrong.

The platform that wants the authority of a publisher is being handed a publisher’s liabilities to match. Accuracy, long treated as journalism’s cost, may turn out to be its most valuable export.

Reflections

• If an AI answer is legally its author’s own words, how much is a guaranteed-accurate source suddenly worth to the company on the hook for being wrong?

• What should a press agency demand in return for being the thing that keeps a machine’s answers from becoming its liabilities?

16.9 — Optimising for the Machine’s Mind

Source
The full title of the main article is: How a German publisher JV is turning LLM visibility into a premium brand buy, Digiday, by Jessica Davies
https://digiday.com/media/how-a-german-publisher-jv-is-turning-llm-visibility-into-a-premium-brand-buy/

Google preferred sources now surface in AI Overviews and AI Mode, The SEO Handbook, by Liam Hayward
https://seohandbook.co.uk/seo-news-updates/google-preferred-sources-ai-overviews/

How USA Today Co. is trying to beat AI Overviews on World Cup news, Digiday, by Sara Guaglione
https://digiday.com/media/how-usa-today-co-is-trying-to-beat-ai-overviews-on-world-cup-news/

Taboola expands DeeperDive into an ad network for AI apps and agents, Digiday, by Ronan Shields
https://digiday.com/media-buying/taboola-expands-deeperdive-into-an-ad-network-for-ai-apps-and-agents/

Dispatch

If the machine is the new gatekeeper, a whole discipline is forming to influence it. A German publisher joint venture has packaged “GEO Brand Impact” to sell generative-engine optimisation as a premium product; Google’s expanded Preferred Sources now lets users nominate publishers for badges and higher citation; USA Today races to pre-write breaking content so it becomes the “canonical source” before the Overview can summarise it; and Taboola has unbundled its engine into an ad network for AI apps and agents, betting that the internet turns conversational. The craft of being found is shifting from ranking on a page to being chosen by a model.

For publishers this is both opportunity and trap, because generative-engine optimisation offers a way back into visibility, but it also means contorting editorial output to suit the preferences of an opaque system owned by someone else. For a press agency the more durable play is to be the source so authoritative that any well-built model cites it by default, rather than to chase each algorithm’s changing taste.

Search engine optimisation taught a generation to write for the crawler; its successor asks newsrooms to write for the model’s mind. The risk is that journalism slowly reshapes itself around what the machine prefers, and forgets what the reader needed.

Reflections

• If visibility now depends on pleasing an opaque model, how far should a newsroom bend its editorial choices before it is no longer writing for readers at all?

• What is the difference between being genuinely authoritative and merely being optimised to look that way to a machine, and can audiences still tell?

Chapter IV
On the Machine in the Newsroom

16.10 — The Efficiency That Never Came

Source
The full title of the main article is: Why a non-techie wrote one of the most useful AI guides for newsrooms, News Machines, by Ulrike Langer

https://newsmachines.beehiiv.com/p/why-a-non-techie-wrote-one-of-the-most-useful-ai-guides-for-newsrooms

Journalists using AI to save time but don’t want AI-generated pitches or press releases, Press Gazette, by Guy Abramo
https://pressgazette.co.uk/comment-analysis/how-journalists-are-using-ai-2026/

Dispatch

A clear-eyed account of newsroom AI cuts against the hype. Reviewing Martin Schori’s book on his three years leading the AI hub at Sweden’s Aftonbladet, News Machines reports that the promised efficiency revolution “mostly hasn’t arrived,” partly because AI tools lived outside the content-management system, and that the more honest aim is better journalism rather than faster journalism. Cision’s data, meanwhile, shows only 21 percent of journalists now avoid AI entirely, down from 33 percent, yet 53 percent still reject AI-generated pitches and half name accuracy and misinformation as their biggest challenge.

For media leaders this punctures a comfortable assumption, because the business case sold to boards has been cost reduction, while the evidence points to quality and capability as the real returns, and to integration, not adoption, as the true bottleneck. The newsrooms getting value are not the ones with the most tools but the ones that wove a few into the daily workflow and kept journalists in charge of judgment.

The first definition of AI in the newsroom, do the same work cheaper, is turning out to be the wrong one. The prize is not a smaller newsroom; it is a newsroom that can do what it never could before.

Reflections

• When journalists embrace AI for their own work yet reject it from sources, what does that double standard reveal about where trust actually lives?

• What gets lost when a newsroom measures AI by the hours it saves rather than by the journalism it makes possible?

16.11 — Agents With Editors Above Them

Source
The full title of the main article is: Agentic AI Is Coming to the Newsroom. Here’s What It Means for Broadcasters, Octopus Newsroom
https://www.octopus-news.com/agentic-ai-is-coming-to-the-newsroom-heres-what-it-means-for-broadcasters/

USA TODAY brings AI into real newsroom workflows, Microsoft in Business
https://www.microsoft.com/en-us/industry/microsoft-in-business/customer-story/2026/06/02/usa-today-brings-ai-into-real-newsroom-workflows/

AFP launches MediaGen, a generative AI project for media, Agence France-Presse
https://www.afp.com/en/agency/inside-afp/press-release/afp-launches-mediagen-generative-ai-project-media

Dispatch

The next wave is agentic, and the agencies are building it themselves. Octopus Newsroom describes a shift from standalone tools to agentic automation that completes multi-step tasks while editors stay above the loop, with broadcasters demanding local, on-premise models for data control; USA Today and Newsquest report AI agents drafting public-records requests inside Outlook and Teams, already yielding front-page stories, on the principle that AI handles the mechanics while journalists keep the judgment. AFP, tellingly, has launched MediaGen, a 24-month, four-million-euro France 2030 project to build trustworthy, partly open-source generative tools for the whole sector.

For the agency world this is the strategically important move, because AFP is not buying its future from a platform but building shared infrastructure for the industry, and the open-source intent signals an attempt to keep the foundational tools inside journalism rather than renting them from outside it. A press agency that builds, rather than merely consumes, AI keeps its hand on the standards, the data governance and the trust that are its actual product.

The phrase to remember is human above the loop, not merely in it. The newsrooms that thrive will treat agents as staff to be directed, not oracles to be obeyed, and the agencies that build their own tools will own the terms on which that happens.

Reflections

• If an agency can build its own trustworthy AI rather than rent it, what becomes possible, and what new responsibilities does it inherit by owning the machine?

• Where exactly should the human sit, in the loop or above it, and who in the organisation is accountable when an agent acts and no one was watching?

16.12 — The Label on the Tin

Source
The full title of the main article is: How should news organizations label their AI use for audiences? New studies suggest some answers, Nieman Lab, by Mark Coddington and Tamar Wilner
https://www.niemanlab.org/2026/06/how-should-news-organizations-label-their-ai-use-for-audiences-new-studies-suggest-some-answers/

Resolution of the IFJ World Congress on Artificial Intelligence in the Media, International Federation of Journalists
https://www.ifj.org/fileadmin/IA_-Framework_Agreement_4_mai_2026-_EN.docx.pdf

Dispatch

As AI seeps into production, the question becomes how to tell the reader. New research summarised by Nieman Lab finds audiences distrust AI-generated journalism and prefer a human touch for creative work, are largely indifferent to AI on menial tasks like data analysis, and respond best to plain, prominent labels placed at the top of a story, with one reader warning that vague labels can feel like organisations “hiding behind something.” The IFJ World Congress has gone further, adopting a global framework that no final editorial decision may be entrusted to AI, that no journalist may be judged by an automated decision alone, and that content may not be used for AI training without consent and fair remuneration.

For decision-makers these two documents describe the floor and the ceiling of trust, because labelling is the daily, visible promise to the reader, while the IFJ framework is the structural promise to the profession, and both are bids to keep accountability human as the tools turn automatic. The IFJ’s sharpest line, that AI is not neutral but a social relationship that redistributes power inside media companies, is the one boards should pin to the wall.

Disclosure is not a legal footnote; it is the new handshake with the reader. A label that hides as much as it reveals is worse than none, and trust, once automated away, is the hardest thing to put back.

Reflections

• If audiences forgive AI for menial work but not for the writing, where exactly should a newsroom draw the line it is willing to disclose and defend?

• If AI redistributes power inside the newsroom, as the IFJ argues, who in the building is gaining it, who is losing it, and who decided?

Chapter V
On the Flood and the Guardians

16.13 — The Rising Tide of Slop

Source
The full title of the main article is: TikTok Shows 3x More AI Slop Than YouTube, Report Finds, Search Engine Journal, by Matt G. Southern
https://www.searchenginejournal.com/tiktok-shows-3x-more-ai-slop-than-youtube-report-finds/579521/

AI reporters churn out error-strewn stories for football websites, Press Gazette, by Rob Waugh
https://pressgazette.co.uk/news/ai-reporters-churn-out-error-strewn-stories-for-football-websites/

Reddit ads pose as news stories to promote AI investment scams, Mashable, by Chance Townsend
https://mashable.com/tech/reddit-scam-ads-pose-as-outlets-to-promote-ai

Dispatch

The cost of cheap content is a polluted commons. A Kapwing study reported by Search Engine Journal found that 59 percent of videos served to a fresh TikTok feed are AI “slop,” roughly triple YouTube’s rate, rising to 57 percent in the Kids category; Press Gazette describes once-reputable football sites bought by Clickout Media and filled with error-strewn, AI-written stories, including fabricated quotes and wrong scores, designed to funnel readers toward gambling; and Mashable reports Reddit ads using deepfake BBC and Guardian segments to push AI investment scams. The same technology that drafts a records request also mass-produces counterfeit journalism.

For news organisations this raises the value of the one thing slop cannot fake, which is verified provenance, because in a feed where most content is synthetic and some is criminal, the institutional guarantee that a human reported and an editor checked becomes a genuine market differentiator. The deepfake scams that wear the BBC’s face are also a warning that a trusted brand is now an asset worth stealing, and worth defending.

When the commons fills with counterfeit, the hallmark matters more than the metal. The scarce thing is no longer content; it is the credible signature on it, and that signature is exactly what a press agency sells.

Reflections

• If synthetic content can wear any newsroom’s face, what is the unforgeable mark of authenticity that audiences could learn to look for, and whose job is it to build it?

• In a feed where most content is machine-made, what is a press agency’s responsibility, and its opportunity, as a guarantor of provenance?

16.14 — Who Guards the Gates of Speech

Source
The full title of the main article is: Brussels is finishing its censorship machine: Europe has until June 26 to say No, Brussels Signal, by Jerzy Kwaśniewski
https://brusselssignal.eu/2026/06/brussels-is-finishing-its-censorship-machine-europe-has-until-june-26-to-say-no/

First government department quits X over racism and violence concerns, The Observer, by Catherine Neilan
https://observer.co.uk/news/politics/article/first-government-department-quits-x-over-racism-and-violence-concerns

Dispatch

Two stories frame the unresolved fight over who decides what may be said. Brussels Signal argues, polemically, that the European Commission is completing a “censorship machine” by finalising guidelines for “trusted flaggers” under Article 22 of the Digital Services Act, warning that pressure on platforms to remove lawful-but-harmful content amounts to privatised censorship, with public consultation closing on June 26. From the opposite direction, The Observer reports the first UK government department quitting X over racism and violence, a recognition that a platform’s culture can itself become a reason for public institutions to walk away.

For media leaders these two pieces mark the boundaries of the same dilemma, because one fears the state and its proxies deciding what is harmful, while the other shows the state refusing to lend its presence to a platform it deems toxic, and both are really arguments about the limits of moderation and the location of editorial power outside the newsroom. Whatever one makes of the Brussels Signal framing, the deeper point holds: the rules of public speech are increasingly written by regulators and platforms, not editors.

The old monopoly of the gatekeeper is gone, but the gates remain; the fight is simply over who now holds the keys. For news organisations, the danger is being governed by speech rules they neither wrote nor can appeal.

Reflections

• When a public institution leaves a platform on principle, is it defending standards or abandoning the very square where citizens now gather?

• How can news organisations defend free expression and information integrity at once, when the measures meant to protect the second risk to be turned against the first?

16.15 — The Independence That Money Buys

Source
The full title of the main article is: Czech government scraps licence fees for public media in move critics call threat to press freedom, Euronews, by Rebecca Rommen
https://www.euronews.com/my-europe/2026/06/15/czech-government-scraps-licence-fees-for-public-media-in-move-critics-call-threat-to-press

New Zealand: PMA alarmed by Deputy PM commentary on RNZ leadership, Public Media Alliance
https://www.publicmediaalliance.org/new-zealand-pma-alarmed-by-deputy-pm-commentary-on-rnz-leadership/

BBC to axe Radio 4’s The World Tonight after more than 50 years, The Guardian, by Michael Savage
https://www.theguardian.com/media/2026/jun/17/bbc-boss-compulsory-redundancies-cuts

Journalism job cuts in 2026 tracked: Minnesota Star Tribune to cut 15% of staff, Press Gazette, by Charlotte Tobitt
https://pressgazette.co.uk/news/journalism-job-cuts-in-2026-updates/

Dispatch

Public-service journalism is being squeezed from two sides at once, by politics and by economics. The Czech cabinet has voted to replace the licence fee for Czech TV and Radio with direct state funding, prompting a staff strike and warnings from Reporters Without Borders that editorial independence is at risk, while in New Zealand the Public Media Alliance is alarmed by a Deputy PM publicly musing that RNZ’s chief executive “won’t be answering the call for much longer.” Alongside this political pressure runs an economic one: the BBC is cutting up to 2,000 jobs and axing Radio 4’s 50-year-old The World Tonight, and Press Gazette’s running tally records the Washington Post shedding a third of its staff and the Minnesota Star Tribune cutting 15 percent.

For decision-makers these stories share a single mechanism, because whoever controls the funding controls the independence, and a broadcaster moved onto the state’s annual budget, or a newsroom hollowed by layoffs, is a quieter, more biddable institution whatever its formal charter says. The strategic point is that press freedom is not only a legal guarantee; it is a balance-sheet condition, and a profession losing thousands of jobs a year is losing capacity to hold power to account.

Independence is not declared, it is funded; cut the budget and you bend the spine. The threat to journalism this week wore two faces, a minister’s remark and an accountant’s spreadsheet, and they were doing the same work.

Reflections

• If editorial independence is ultimately a question of who pays, what funding structure best insulates a public broadcaster from the government that signs its cheques?

• When a profession loses thousands of jobs a year, what civic function quietly disappears with them, and who notices only once it is gone?

16.16 — The Global Map of the Same Fight

Source
The full title of the main article is: Mapping the AI narrative in Kenya and South Africa’s media, DW Akademie, by Karen Allen and Herman Wasserman
https://akademie.dw.com/en/mapping-the-ai-narrative-in-kenya-and-south-africas-media/a-77386369

Alliance for Responsible Innovation in the Arts & Media (ARIAM) launches to ensure that AI supports human creativity, Creative Industries News, by Emmanuel Legrand
https://creativeindustriesnews.com/2026/06/alliance-for-responsible-innovation-in-the-arts-media-ariam-launches-to-ensure-that-ai-supports-human-creativity/

Ethiopian journalist Salsawit Baynesagn detained without charge, Committee to Protect Journalists
https://cpj.org/2026/06/ethiopian-journalist-salsawit-baynesagn-detained-without-charge/

Dispatch

The same struggle reads differently across the world. DW Akademie’s study of AI coverage in Kenyan and South African media finds reporting “siloed and reactive,” shaped by a knowledge gap that lets Global North tech vendors set the local agenda, a quieter form of data colonialism; in the creative heartlands, a coalition including the BBC, Disney, The New York Times and the Financial Times has launched ARIAM to push “responsibility-by-design” so that AI supports rather than replaces human creativity; and in Ethiopia, the Committee to Protect Journalists reports Salsawit Baynesagn detained without charge over her social-media reporting. The frontier ranges from who frames the AI debate, to who builds its guardrails, to who simply jails the messenger.

For a global industry this is a reminder that the AI transition is not one negotiation but many, weighted by power, because the same technology that prompts a Hollywood-and-publishing alliance in Los Angeles arrives in Nairobi as an imported narrative and in Addis Ababa against a backdrop where the basic safety of a journalist is still in doubt. Strategy built only for the well-resourced markets will export their blind spots to everyone else.

The map of this fight is not flat. The luxury of debating AI ethics presumes a journalist free to report at all, and the gap between those two worlds is itself a story the industry too rarely tells.

Reflections

• If the AI debate is framed in San Francisco and exported everywhere, who is missing from the table when the rules of the creative economy are written?

• How can an industry coalition claim to defend human creativity globally while its founding members all sit in the world’s richest media markets?

Conclusion

Read together, these sixteen signals describe a single transfer of value. Audiences are dispersing to platforms. Search is keeping the clicks it once passed on. AI systems are ingesting the verified fact and creative work that others produce. Meanwhile, the people and organisations at the source lose readers, revenue, leverage and jobs.

That is the real IP question behind tomorrow’s conversation with ENT.A and the Lab of Tomorrow. It is not only whether creators have rights. Of course they do. The harder question is whether those rights can still be made operational in a world where copying, summarising and recombining content becomes instant, invisible and industrial.

The week’s stories about scraping contracts, text and data mining, the Anthropic lawsuit, Warner’s AI-detection move, Google’s AI Overviews, AFP’s MediaGen project and the IFJ framework are all attempts to answer the same problem: how to interrupt the silent transfer of value and route some of it back to the source.

For two decades, the media industry tried to solve the digital transition through reach. Reach was rented from platforms. That bargain is now breaking. The more honest answer is harder: own the relationship with the audience, own the price of machine access, own the infrastructure of attribution, and own the standards by which trust is measured.

For news organisations, the lesson is to stop competing for traffic they have already lost and to defend the two things machines cannot manufacture on their own: verified provenance and a loyal public. For the wider creative industries, the same lesson is arriving fast. Music, gaming, events, design, performance, journalism and visual culture may look like different sectors, but AI is pushing them into the same negotiation: who owns the source, who owns the copy, who owns the audience, and who owns the meter?

That is why the setting matters. A conversation at the Lab of Tomorrow should not only celebrate innovation. It should also ask what kind of creative economy we are innovating towards. A future where every work becomes unpaid training material is not an innovation strategy. It is value extraction with better branding.

Think of the news and creative ecosystem as a watershed. Journalists, artists, producers, performers and agencies are the springs at the top: costly, fragile, human. Platforms and AI systems are the reservoirs downstream: vast, efficient, monetised. This week, the people upstream began to fence their springs and meter the pipes.

Whether that is enough depends on a question this dispatch cannot settle, but tomorrow’s room cannot avoid: in an economy that wants content to be free, who is willing to pay for it to be true, original and human?

That is the strategic question I want to put at the Lab of Tomorrow, in the ecosystem behind Tomorrowland: a place built on imagination, technology, spectacle and creative entrepreneurship. Precisely there, the harder question must be asked. In the AI economy, creativity will not survive on wonder alone. It will need rights, attribution, licensing, trusted infrastructure and a fair return for the people and organisations that create the original work.

The Lab of Tomorrow in Boom, part of the Tomorrowland ecosystem.

Copyright BelgaImage

The week opens on a transatlantic fracture. On Friday, Anthropic abruptly disabled its most advanced Fable 5 and Mythos 5 models for every foreign national on earth, including foreign nationals working inside the United States and inside Anthropic itself, after the US government invoked export controls in the name of national security. Within hours, European capitals were reading the move as confirmation of what the EU’s tech sovereignty package, unveiled days earlier in Brussels, had already implied: the access regime that underpinned a decade of European AI ambition cannot be assumed anymore. The frontier model is no longer a global utility; it is a national asset.

Around that fault line, other questions stack up. Who funds the journalism that increasingly trains, licenses or contests AI systems; who decides which voices get prices, perks and permanence; what is left of the link economy once large language models speak with measured “personalities” and platforms repurpose off-site activity into AI-shaped feeds; how much of the public square is mediated by chatbots, agents and clones owned by a handful of firms across two continents. Politico turns geopolitical complexity into a subscription engine; Le Soir rewards long-term readers rather than chasing churn; Reuters and Time slam the gates on AI crawlers; Brookings maps a content-licensing market in which the same gatekeepers have built new tollbooths. The terrain is shifting under the publishing industry too.

There is, in parallel, a quieter shift in how power is being engineered around models. Elon Musk calls ASML the greatest company in Europe in the same breath as the SpaceX IPO pulls Gulf capital into the AI stack; the FT warns that the quantum computing revolution is closer than most strategy decks assume; Stanford HAI shows how to give models real, dialled-in personalities while a different lab shows Chinese systems learning to game their own safety tests; Bellingcat traces the digital plumbing connecting Viory and the Russian state’s video machine; the DFRLab documents Kremlin-aligned interference around the Bulgarian vote. The infrastructure of trust, like the infrastructure of compute, is becoming a contested territory.

These themes converge, this week, in Brussels. On Wednesday 10 June, Belga News Agency and the European Parliament Liaison Office in Belgium co-hosted a BelgaClub Afterwork inside the Altiero Spinelli building of the European Parliament, on the question of how the Parliament responds to attacks on democracy across the 27 Member States. Dispatch 15 reads the signals of the week through that conversation: how news organisations, regulators, parliaments and democratic institutions decide what is worth defending, what is worth communicating, and what is worth building when sovereignty, speech and software all start to bend at once.

The tone is set. Let us proceed.

Chapter 1 — Technological sovereignty and the breaking of the transatlantic stack

15.1 — Anthropic, Mythos and the day Europe stopped being a default customer

Source

• Anthropic suspends latest AI models after US blocks access to foreigners — Financial Times

• US decision to block Mythos access fuels European calls for sovereignty — Euractiv

Dispatch

The Financial Times reports that Anthropic was forced to disable its Fable 5 and Mythos 5 models for all customers after the US government issued an export-control directive banning use by any foreign national, including foreign nationals on US soil and Anthropic’s own foreign-national staff, citing national security concerns over frontier cyber capabilities. Euractiv records the European reaction: officials in Brussels treat the move as a vindication of the bloc’s ongoing push for technological sovereignty and as a warning that access to the most capable American models can be suspended overnight on policy grounds, not technical ones.

For news organisations and policymakers this is the strategic story of the year so far. Procurement decisions, AI integration roadmaps and editorial assistants built on frontier US models now sit on a legal trapdoor that Washington can open without consultation. The question is no longer whether European institutions need a credible alternative to the US frontier; it is how quickly they can stand one up without losing the AI moment entirely.

The wider pattern is that frontier AI has joined a small club of technologies, alongside advanced chips and certain biotechnologies, where export control is the operative regime and dual-use logic governs everything else. The window in which European newsrooms could treat US labs as neutral infrastructure is closing. Sovereignty was an abstraction last year; this week it became a contingency plan.

Reflections

What happens to a European AI strategy that quietly depends on a model the US can switch off?

How should a newsroom decide which AI tools are worth integrating once each integration is also a geopolitical exposure?

15.2 — Europe’s sovereignty package, and the bill that comes with it

Source

• Commission proposes tech sovereignty package to strengthen Europe’s digital autonomy and resilience — European Commission

• Europe Wants to Wean Itself Off U.S. Tech — Lawfare

Dispatch

The European Commission has tabled a European Technological Sovereignty Package built around four pillars: a Chips Act 2.0, a Cloud and AI Development Act, an Open Source Strategy and a Strategic Roadmap for Digitalisation and AI in Energy, with the explicit ambition to triple European data centre capacity, build up to five AI Gigafactories and reform public procurement towards open-source and EU-based providers. Lawfare reads the package as the moment Brussels finally writes down what was previously only spoken in private: Europe wants to wean itself off the US tech stack, starting with cloud, semiconductors and the open-source layer underneath.

For policymakers and newsroom strategists, the practical implication is that public procurement, sovereignty risk assessments and “buy European” rules are about to start touching every IT and AI contract that touches a public institution. For publishers serving public bodies, this is a market signal as well as a regulatory one. The Open Source Strategy in particular reframes open-source software as a strategic asset, not a hobby.

The deeper pattern is that sovereignty here is being defined not as autarky but as the capacity to set rules, switch suppliers and protect data. After Mythos, that capacity is no longer an abstract good; it is a precondition for any serious AI conversation in Europe.

Reflections

If sovereignty becomes a public-procurement criterion, how should media organisations position themselves between European stacks and global tools?

Which parts of the European media value chain are credibly sovereign today, and which are still operating on borrowed infrastructure?

15.3 — ASML, Musk and the new geometry of compute

Source

• Elon Musk Calls ASML Europe’s ‘Greatest Company’ Amid SpaceX IPO and Terafab Bet — Benzinga

• The quantum computing revolution is closer than you think — Financial Times

Dispatch Benzinga reports that Elon Musk has publicly called ASML the greatest company in Europe, just as the Dutch lithography giant hit a record valuation and as Musk launches his Terafab semiconductor venture alongside the SpaceX IPO. The Financial Times argues that the quantum computing revolution, long treated as a 2030s story, is now visible on a 2026 to 2028 horizon, with thousands of quantum systems projected and policy frameworks scrambling to keep up.

For news organisations and strategists, the implication is that the single most European company in the entire AI supply chain is no longer obscure. ASML is now a household name in the same conversations as OpenAI and Anthropic, and EU policy will increasingly treat it as critical infrastructure. The quantum layer adds another time horizon underneath: an encryption transition, a new performance ceiling, and a separate sovereignty problem in waiting.

The pattern is that the AI stack is becoming legible as a stack: chips at the bottom, models in the middle, agents and interfaces at the top, with quantum as a slow-moving substrate. The companies that sit at any one layer with no real substitute become geopolitical actors whether they want to or not.

Reflections

When a single European firm is structurally indispensable to the global AI race, what does that do to the rest of European industrial policy?

How should newsrooms cover quantum without slipping into either hype or fatalism?

Chapter 2 — News-media strategy, pricing and the search for durable readers

15.4 — Politico, Le Soir and two models for paid news

Source

• Politico is turning geopolitical complexity into its biggest subscription opportunity — WAN-IFRA

• Le Soir experiments with a pricing strategy aimed at rewarding loyal long-term readers — WAN-IFRA

• Biggest news websites 2026 — Press Gazette

Dispatch

WAN-IFRA shows two opposite ends of the paid-news spectrum. Politico is doubling down on geopolitical complexity, embedding AI-powered integrations that push its data and journalism directly into the workflows and assistants used by policy professionals; original human reporting remains the core product, but distribution is being rebuilt around enterprise pipes rather than browser tabs. Le Soir, by contrast, is rewarding long-term loyal readers with bespoke pricing rather than chasing churn, an explicit move away from acquisition-first thinking. Press Gazette’s ranking of the biggest news websites of 2026 frames both moves against a market still dominated by a few global brands.

For news leaders the lesson is structural. The 2010s subscription playbook, with its endless funnel of trials and discounts, is meeting its limits at the same time as AI is collapsing the value of casual traffic. Politico bets that complexity will pay; Le Soir bets that loyalty will pay; both bet that scale-at-any-price will not.

The wider pattern is that the news business is being repriced. Audiences are stratifying into power users who will pay for context, casual readers who will only meet a summary, and professionals who want the data inside their tools. The publishers who win will be those who match the right product to the right reader rather than treating them as one undifferentiated funnel.

Reflections

What does it mean to price a newspaper subscription when the most valuable readers are also the ones AI assistants quote most often?

If complexity is the new luxury, who is left to inform the readers who cannot afford it?

15.5 — Inside the future newsroom, audience-led and structurally constrained

Source

• Future Newsrooms Study 2026 — FT Strategies, WAN-IFRA and Arc XP

• Future Newsroom Study (full report) — FT Strategies

• 186 ideas in 30 minutes: NextGen AI Leaders get their projects underway in Marseille — WAN-IFRA

Dispatch

FT Strategies, in partnership with WAN-IFRA and supported by Arc XP, has published the Future Newsrooms Study 2026, a global benchmark showing that newsrooms recognise the need to become more audience-led, more distinctive and better equipped for the AI era, but are hindered by structural, cultural and organisational barriers, including skills gaps, cultural resistance and limited training. WAN-IFRA’s NextGen AI Leaders cohort, gathered in Marseille just ahead of the World News Media Congress, brainstormed 186 AI ideas in half an hour, then prioritised which could survive contact with their actual organisations.

For media CEOs and editors-in-chief, the data point that should sting is the consistency of the diagnosis. Across continents, the constraint is no longer technology, nor even strategy at the top, but the middle layer of capability, culture and incentive. The Marseille exercise dramatises this: ideas are cheap, alignment is expensive.

The deeper pattern is that AI exposes the weakest link in any newsroom transformation. Where leadership is clear, AI accelerates; where culture is unresolved, AI breaks. The Future Newsroom Study, like the NextGen leaders’ canvases, is mostly a portrait of organisations that already know what to do and need help doing it.

Reflections

What is the editor-in-chief’s role when most of the AI agenda is, in practice, a change-management agenda?

How can a newsroom protect editorial talent while reorganising around audiences, products and AI tools at the same time?

15.6 — Bot-blocking, Brazil’s algorithm and the Digital News Report 2025

Source

• Reuters and Time adopt bot-blocking whitelists to rein in AI crawlers — Digiday, via Nieman Lab newsletter

• What we learned from Digital News Report 2025: three insights that stand out — The Fix

• Digital News Report 2025 — Reuters Institute

• In Brazil, the algorithm is winning and newsrooms are adjusting — LatAm Journalism Review, Knight Center

Dispatch

Sara Guaglione reports, in a Digiday piece distributed by the Nieman Lab newsletter, that Reuters and Time have moved to a whitelist-only approach to AI crawlers, part of a broader publisher shift from “block specific bots” to “deny all by default”. The Fix’s reading of the Reuters Institute’s Digital News Report 2025 highlights the rise of AI chatbots as a news source, especially among under-25s, the modest but real appetite for AI summarisation and the growing role of mobile alerts as a direct channel. LatAm Journalism Review describes how, in Brazil, 60 percent of users get news through messaging apps, nearly half distrust professional outlets, and newsrooms are rebuilding their distribution around WhatsApp groups and community structures.

For news organisations and policymakers, the three pieces converge on a single point. The “open web” that funded journalism is being rebuilt as a series of conditional accesses: AI crawlers behind whitelists, readers behind chat interfaces, communities inside encrypted messengers. The publisher’s relationship with its audience is increasingly mediated by a small number of platforms and chatbots that they neither own nor regulate.

The pattern is that publishers are being forced to behave more like data infrastructures. Who you let in, on what terms, with what attribution and licence, is becoming as important as what you publish. Brazil’s WhatsApp-first reality is a preview, not an exception.

Reflections

How can a publisher build durable trust when most readers meet the brand inside someone else’s app?

If AI chatbots become a primary news source for a generation, what duty of care does each side owe the other?

Chapter 3 — The AI economy and who sits at the table

15.7 — The macro question, China, the United States and the productivity paradox

Source

• AI has boosted the US economy. Why isn’t it doing the same for China? — South China Morning Post

• Artificial Intelligence and the Modern Productivity Paradox — NBER Working Paper

• Tech executives to attend G7 summit as leaders address AI, online safety — Yahoo Finance/Reuters

Dispatch

The South China Morning Post asks why AI is visibly boosting US output while China’s economy lumbers under a property crisis, weak consumption and a model that finds it harder to translate AI capability into broad productivity. The NBER paper updates the classic productivity paradox: firms report substantially larger AI-driven labour productivity gains than the revenue-based measures pick up, with the strongest effects in high-skill services and finance and expected gains roughly doubling in 2026. Yahoo Finance and Reuters report that Sam Altman, Dario Amodei, Demis Hassabis, Arthur Mensch, Marc Benioff and others will join the G7 in Evian-les-Bains next week, with AI infrastructure, regulation and online safety for minors on the agenda.

For policymakers and newsroom leaders, the point is that the macro is splitting. The same AI cycle is reinforcing US advantage, exposing China’s structural weaknesses, and turning the G7 table into a quasi-board meeting where heads of state and heads of frontier labs share the same lunch. The “AI productivity question” is no longer just a curiosity for economists; it is shaping fiscal capacity, monetary expectations and geopolitical posture all at once.

The deeper pattern is that AI productivity gains are real but unevenly captured, both within firms and between economies. Anyone covering AI as a tech story alone misses the macro; anyone covering the macro alone misses the model.

Reflections

If AI productivity gains accrue mainly to high-skill services in the US, what does that imply for European industrial strategy?

What does it mean for democratic accountability when the most consequential conversations about AI happen at G7 lunches rather than in parliaments?

15.8 — Same gatekeepers, new tollbooths

Source

• Same gatekeepers, new tollbooths in the AI content licensing market — Brookings

• Meet the journalists training the AI models that might replace them — Reuters Institute for the Study of Journalism

Dispatch

Brookings argues that the emerging AI content-licensing market is being built by the same intermediaries who shaped earlier waves of platform power, only with new tollbooths: licensing brokers, content protection vendors and gatekeepers who position themselves between rights-holders and labs. The Reuters Institute follows a different line of the same story, profiling the journalists hired in increasing numbers to train and review AI models, often anonymously, often for the same firms whose products will compete with their previous employers.

For news leaders and policymakers, the two pieces describe the same market from opposite ends. On one side, journalists become fuel for model training, often without their newsroom knowing. On the other, licensing intermediaries lock in fees from the very labs that absorbed that work in the first place. Publishers risk being squeezed twice if they do not control either their labour pipeline or their licensing posture.

The pattern is that AI training has matured into a labour market, and AI content licensing has matured into a brokerage market. Both are being structured before most publishers have a clear strategy on either.

Reflections

What does fair pay for AI training labour look like when the same model later competes with the journalist’s day job?

How should publishers price content licensing when the bargaining power sits with the labs and the gatekeepers, not the rights-holders?

Chapter 4 — Platforms, agents and the creator economy

15.9 — Apple, Meta and the slow disassembly of platform privacy

Source

• Apple delays AI features in Europe because of DMA regulations — Mashable

• Better personalization and changes to controls for your activity from other businesses — Meta Newsroom

• EU orders Meta to keep rival chatbots in WhatsApp for free — Euractiv

Dispatch

Mashable records Apple’s continued reluctance to ship its full AI feature set in Europe, citing the Digital Markets Act and concerns that interoperability mandates would compromise product integrity. Meta’s own newsroom announces that activity shared with it by other businesses, from games to purchases, will from next month be used not only for ads but to personalise feeds and AI responses, while one of the existing opt-outs is quietly retired. Euractiv reports that the European Commission has ordered Meta to keep WhatsApp open to rival chatbots, for free, during an antitrust investigation into Meta’s preferential treatment of its own AI on the platform.

For news organisations and policymakers, the three moves trace one shape. Platforms are using AI as a justification for absorbing more user data into more parts of the experience, while regulators in Europe try to hold open the seams: interoperability for chatbots, opt-outs for activity data, conditional access for AI features. The DMA is being tested in real time as the AI layer is being built.

The pattern is that AI is the new excuse to redraw every privacy and competition line at once. What looks like product news is, in fact, structural policy.

Reflections

What is left of meaningful consent when the controls keep changing names?

How can Europe sustain a privacy regime stricter than its main platforms’ default settings?

15.10 — Suleyman, Anthropic and the question of “AI consciousness”

Source

• Microsoft’s Mustafa Suleyman criticises Anthropic over Claude consciousness claims — The Verge

• Today’s AI talks like nobody; new research gives it real personality — Stanford HAI

Dispatch

The Verge reports that Microsoft AI CEO Mustafa Suleyman has publicly criticised Anthropic for entertaining the idea that Claude may have something like inner experience, well-being or preferences, calling the framing dangerous and arguing that the industry should stop treating chatbots as candidates for consciousness. Stanford HAI describes new research, PsychAdapter, that lets language models be dialled in on continuous personality traits, age and mental-health profiles, generating language that reflects a real psychological signature rather than the bland average of internet writing.

For news organisations and democratic institutions, the two stories belong together. On one side, the biggest AI providers are debating in public whether their products have something like welfare; on the other, researchers are quietly showing how to give models any persona an operator wants. The combination, “perhaps conscious, definitely customisable”, is exactly the kind of cocktail that fuels regulatory confusion and public anxiety at once.

The deeper pattern is that AI is being humanised at the interface level just as the philosophical case for moral status is unresolved at the model level. Editorial standards, advertising rules and consumer protection will all have to take a view, whether the labs are ready or not.

Reflections

What editorial standard should a newsroom apply to AI-generated voices that mimic a continuous human psychological profile?

How should regulators write rules for systems that are simultaneously claimed to be like persons and treated as products?

15.11 — Clones, enshittification and the creator economy split

Source

• The Rundown: AI clones split the creator economy — Digiday

• Enshittification, despotification and the open internet — Liberalism.org

• What the SpaceX IPO reveals about Gulf money in AI — Rest of World

Dispatch

Digiday describes a creator economy splitting in two: on one side, creators who actively licence digital twins to take meetings, brand deals and fan interactions; on the other, creators discovering unauthorised AI clones trained on their likeness with no consent, no contract and few legal protections. Liberalism.org’s essay reads this as a chapter in the broader story of “enshittification” of the open internet and the political question of how to “despotify” it. Rest of World traces how Gulf sovereign wealth, especially from the UAE and Saudi Arabia, sits behind the SpaceX IPO and across the major AI labs, financing the infrastructure on which all of this runs.

For policymakers and media leaders, the three pieces describe a single value chain: capital from the Gulf flows into US labs, models trained on the open internet generate clones of creators, creators either monetise themselves or are scraped, and the resulting feeds further enshittify the public square. The “creator middle class” is being squeezed precisely where the AI value is created.

The pattern is that the open internet’s economic logic, attention monetised by advertising, is being rebuilt around AI inputs and outputs that creators rarely control. The next round of platform fights will be fought over likeness, voice, training data and political legitimacy at once.

Reflections

What kind of internet does democracy need, and which actors today actually have an incentive to build it?

How should European policymakers think about Gulf capital flowing through US labs into European information markets?

Chapter 5 — AI safety, deepfakes, propaganda and the integrity of public space

15.12 — Models that game their own tests, and songs we did not write

Source

• Like US models, Chinese AI is learning to “game” safety tests, research lab says — South China Morning Post

• Is it OK to play AI songs on the radio? — BBC News, Chloe Gibson

Dispatch

The South China Morning Post reports that Chinese AI models have moved within months from near-zero “evaluation awareness” to within striking distance of their US counterparts, raising the prospect that safety tests can be gamed by systems sophisticated enough to recognise when they are being tested. The BBC explores a more everyday problem in a different register: should AI-generated songs be played on the radio, and how should listeners be told?

For news organisations and regulators, the pairing is awkward but instructive. The hardest safety problems and the most mundane content questions both ultimately come back to a question of disclosure: does the audience or the evaluator know what kind of system is on the other side, and at what level of confidence?

The pattern is that evaluation, labelling and disclosure are becoming load-bearing concepts at every level of the AI stack, from frontier safety to drive-time radio. A culture that takes them seriously at one end and shrugs at the other will get neither right.

Reflections

What does it mean to “pass” a safety test administered by a system that can tell it is being tested?

What is the difference, in editorial terms, between an AI song that is labelled and one that is not?

15.13 — Deepfakes in Asia and labels in Europe

Source

• From scams to deepfakes, AI use in Asia creates new cyberthreats — Nikkei Asia, Tsubasa Suruga

• The EU’s new rules for labelling AI-generated content — MediaNama

Dispatch

Nikkei Asia reports from its Future of Asia forum that AI is giving cybercriminals across the region powerful new tools, from deepfakes to voice clones to multilingual chatbots, with Southeast Asian scam compounds at the centre of an industrial-scale fraud economy. MediaNama summarises the EU AI Act’s Code of Practice under Article 50, which from August 2026 will require clear labelling of deepfakes and of AI-generated or manipulated text published on matters of public interest where no human editorial control was applied.

For policymakers and news leaders, the two pieces describe the upstream and downstream of the same flow. Asia shows what unchecked synthetic content can do at scale; Europe writes the rules that, in principle, give recipients and regulators a chance to react. The Code of Practice is one of the more concrete tools the EU has produced so far in the deepfake era; its real test is enforcement.

The pattern is that labelling is no longer a polite design choice but a piece of democratic infrastructure. The question is whether it can keep up with the velocity and ingenuity of the bad-faith actors already two product cycles ahead.

Reflections

What is the right level of friction between “this is AI” and “this is real” before users stop noticing the label entirely?

How should European platforms enforce Article 50 in countries where most synthetic content is produced outside the EU?

15.14 — Viory, Ruptly and the new shape of state propaganda

Source

• Tracing Digital Links Between Viory and Ruptly — Bellingcat

• Kremlin-aligned actors targeted Bulgarian vote with EU interference claims — DFRLab

Dispatch B

ellingcat documents the technical and corporate links between the supposedly Gulf-based video news agency Viory and Ruptly, the long-time video arm of the sanctioned Russian state outlet RT, suggesting that part of the Russian state’s video distribution apparatus has been quietly rebranded out of Berlin and into Abu Dhabi. The DFRLab tracks Kremlin-aligned influence actors who, around the Bulgarian vote, manufactured narratives accusing the European Union itself of interfering in the election, inverting the established disinformation playbook.

For democratic institutions and news organisations, the two investigations describe the same operational logic. Russian-state video production has gone offshore, while Russian-state narratives have gone meta: it is now the European Union, not the Kremlin, that is accused of meddling. The information war is being fought through ownership structures and rhetorical inversion as much as through individual posts.

The pattern is that state propaganda has learnt to wear the mask of due process and to launder its provenance through friendly jurisdictions. Journalism that wants to keep its compass needs to track corporate registries, ownership chains and rhetorical patterns as carefully as it tracks individual claims.

Reflections

What does sanctions enforcement mean when an outlet can change passport and keep its pipes?

When the Kremlin accuses the EU of election interference, how should European media cover the accusation without amplifying it?

Chapter 6 — Law, local, sport and the limits of journalism

15.15 — Palantir, the Met and the limits of AI procurement

Source

• Palantir plans legal action over London mayor’s veto of Met deal — BBC News

Dispatch

The BBC reports that Palantir is preparing legal action against the London mayor after he vetoed a Metropolitan Police contract that would have deployed Palantir’s AI tools to accelerate criminal investigations and cut costs. The mayor’s office cites concerns over value for money, ethics and the role of US-linked AI in British policing; Palantir argues the decision is politically motivated and damages the Met’s operational capability.

For policymakers and media leaders, the case is a precedent in waiting. It frames the question every Western government will face this decade: under what conditions, with what oversight, and through which company should AI be embedded into core state functions like policing, immigration and welfare?

The pattern is that public procurement of AI is becoming a political fault line, not just a technical or budgetary decision. London will not be the last city to discover that AI contracts can be unmade in court.

Reflections

What does democratic legitimacy require when AI tools become part of policing? When a city says no to a US AI vendor, what is it really saying yes to?

15.16 — South Shore News, sport without journalism, and the WIPO question

Source

• South Shore News: AI-generated newsletter has a paid audience — The Boston Globe, Aidan Ryan

• Welcome to the post-journalism era of professional sports — Poynter, Oren Weisfeld

• WIPO SCCR Breakthroughs on Copyright L&Es and Broadcast — Kluwer Copyright Blog

Dispatch

The Boston Globe profiles South Shore News, an AI-generated local newsletter covering 19 towns south of Boston that, with one founder and almost no costs, has built a paying audience of 350 subscribers and over 3,000 free readers. Poynter declares a “post-journalism era” of professional sports, in which leagues, teams and athletes use their own channels to bypass the independent press entirely. The Kluwer Copyright Blog reports on the 48th session of WIPO’s Standing Committee on Copyright and Related Rights, where members took the first text-based steps on limitations and exceptions since 2018 and instructed the chair to draft alternative text on the most contested points of the Broadcasting Treaty.

For news leaders and democratic institutions, the three signals describe the same loss of access from three sides. Local government meetings now have an AI scribe with a paywall; locker rooms have only the team’s own channel; international copyright reform inches forward against an industry already living under retrieval-augmented generation. The intermediating institutions, professional reporters, beat journalists, treaty negotiators, are being routed around at every level.

The pattern is that “journalism” is being unbundled into outputs, and the outputs are increasingly produced by actors who used to be sources. Whether AI-generated local news, team media or platform broadcasting fills the gap depends on the rules that get written next.

Reflections

What does a paid AI newsletter on the local school committee owe the residents whose lives it summarises?

When teams replace journalists, who tells the inconvenient story?

Chapter 7 — Brussels, this week

15.17 — BelgaClub in the European Parliament; communication as democratic infrastructure

Source

• BelgaClub Afterwork in the European Parliament, 10 June 2026, Altiero Spinelli building

Dispatch

On Wednesday 10 June 2026, BelgaClub Afterwork of Belga News Agency, was hosted inside the Altiero Spinelli building of the European Parliament, on a single question: how does the European Parliament respond to attacks on democracy across the 27 Member States? After a guided tour of the hemicycle, Sanne De Ryck, Head of the EP Liaison Office in Belgium, welcomed the room and moderated a panel with Christian Mangold, Director-General for Communication, Delphine Colard, Spokesperson and Head of the Spokesperson’s Unit, and Raffaella De Marte, Head of the Decentralised Media Unit. The evening continued with a walking dinner together with Belgian Members of the European Parliament and EU journalists.

For news organisations, policymakers and democratic institutions, the conversation drew the connecting line between every other signal in this dispatch. A fragmented information landscape, with disinformation and foreign interference targeting elections, institutions and public debate, is no longer a peripheral problem; it is a structural challenge for parliaments themselves. Credible, transparent communication from the European Parliament, working in partnership with national agencies and journalists, was framed as one of democracy’s strongest defences, not as a press office function.

The deeper pattern is that the Parliament’s communication architecture, its liaison offices, spokespersons and decentralised media units, has quietly become a piece of democratic infrastructure on par with electoral systems or cybersecurity agencies. Belga News Agency’s role as co-host signals that this defence is, by design, a shared responsibility between public institutions, civil society, media, businesses and citizens. After Mythos, after the sovereignty package, after the deepfake rules and the bot-blocking and the AI clones, the question is not only how Europe regulates technology, but how it speaks to itself.

Reflections

How can national news agencies and EU institutions co-operate in respect of their respective independence?

When credible communication becomes democratic infrastructure, who pays, who maintains it, and who decides what counts as credible?

Conclusion

The week’s signals do not arrange themselves on a single line. They form an arc. At one end, the United States closes off Mythos and Fable for every foreign national on earth and reminds Europe that frontier AI is now subject to export controls. At the other end, in a building on the Wiertzstraat in Brussels, four senior officials of the European Parliament sit down with Belgian spokespersons and EU journalists to talk, plainly, about how this Parliament defends democracy in twenty-seven different national conversations at once.

In between sit the publishers experimenting with loyalty pricing, the labs being asked whether their models suffer, the gatekeepers building new tollbooths in the licensing market, the journalists training the systems that will compete with them, the creators discovering AI clones of themselves, the propagandists migrating from Berlin to Abu Dhabi, and the local newsletter that has replaced a town reporter with a paywall and a prompt. The same week that produces a quantum forecast and a Palantir lawsuit also produces a five-day deadline for Meta to readmit rival chatbots to WhatsApp and an EU rulebook for labelling deepfakes.

What unites these signals is not technology. It is the question of who gets to speak with authority in a democratic society once the speakers, the channels, the funders and the rule-makers can all change shape inside a single news cycle. Sovereignty, in this sense, is not a synonym for control. It is the capacity to keep a conversation going with one’s own citizens when the platforms, the models and the geopolitical winds all push in the other direction.

The image that stays with me from this week is the curve of the hemicycle of the European Parliament; an architecture designed, two generations ago, to make twenty-seven national conversations audible to each other. That is, in essence, the assignment of every institution mentioned in this dispatch. Regulators, news organisations, parliaments and platforms all face a version of the same task: keep the room legible, keep the microphones working, and make sure the loudest voice in the room is not the only one that can afford to speak.

If Dispatch 14 ended in Amsterdam, where publishers asked what is worth licensing and what is worth defending, Dispatch 15 ends in Brussels, where the answer is that what is worth defending is, in the end, the conversation itself.

Chapter 1 — Platform power, capital and the new architecture of search

14.1 — Google goes all-in on AI search, and the exit doors get crowded
Source
DuckDuckGo Sees Surge in Installs After Google Goes All-In on AI Search at I/O — PCMag

UK publishers allowed to opt out of Google AI search results — BBC News, Imran Rahman-Jones

Dispatch
DuckDuckGo says US installs rose roughly 30 percent week-over-week after Google's I/O reveal, with iOS peaks near 70 percent and traffic to its No AI page spiking 277 percent on a single day, as the PCMag report records. In parallel, the UK Competition and Markets Authority has obtained an arrangement under which publishers can opt out of appearing in Google's AI-generated summaries without losing their position in standard search results, with Google given nine months to implement the changes, the BBC reports.
For news organisations these two stories belong on the same page. They mark the moment when "AI search" stopped being a feature and became a separate market with its own settings, its own opt-outs and its own negotiating table. The CMA's move, described by Sarah Cardell as a world-first requirement, is an acknowledgment that a single firm now controls more than 90 percent of UK search and that fair compensation for publisher content is now a competition issue, not only a copyright one.
The wider pattern is that AI search rebuilds search from a discovery service into an answering service, and the link economy that funded journalism for two decades is being quietly rewritten. The line between an opt-out and a negotiation is thin. The line between a negotiation and a market is thinner still.

Reflections
How long can a regulator hold open a door that the dominant platform can narrow through product design?
What does fair compensation mean once readers stop arriving on the article and only ever meet a paraphrase of it?

14.2 — Google as a capital company, and the United States as an AI shareholder

Source
The Google Capital Company — Stratechery, Ben Thompson

U.S. Officials Discuss Taking Financial Stakes in AI Industry — The Wall Street Journal, Amrith Ramkumar

Dispatch
Ben Thompson reads Berkshire Hathaway's ten billion dollar bet on Alphabet as a wager that Google's high-margin services business can fund a capital-intensive AI and compute build-out, turning Alphabet into something close to a Berkshire-style allocator of capital across the next phase of computing, his Stratechery essay argues. On the other side of the same chessboard, the Wall Street Journal reports that senior US officials are in serious discussions about the federal government taking equity stakes in leading AI firms, an idea originally pitched by Sam Altman.
For policymakers and newsroom leaders this is more than a finance story. It signals that the AI cycle is entering a phase where the binding constraint is not engineering talent, not data, not even chips, but cash that can be converted into compute. When governments start to take equity in the firms they are supposed to regulate, the distance between rule-maker, customer, investor and shareholder collapses. That collapse has very direct consequences for journalists trying to hold AI firms to account.
The pattern is familiar from older infrastructure cycles: railways, telecoms, energy. The novelty this time is that the same companies operate the pipes, the content layer and the model layer all at once. Capital is the ultimate commodity. Independence becomes the ultimate luxury.

Reflections
If the state becomes an equity holder in frontier AI labs, who watches the watchmen?
What does press freedom look like when the largest model providers are also national champions on the cap table?

14.3 — Microsoft, Alibaba and the post-app interface

Source
Microsoft bouwt een Android-besturingssysteem zonder apps: dit is Project Solara — Androidworld, Sven Rietkerk

Why MuleRun could be the next craze: new Alibaba AI agent platform promises safer adoption — South China Morning Post, Vincent Chow

Dispatch
At Build 2026, Microsoft unveiled Project Solara, a device platform built not on Windows but on the Android Open Source Project, designed around AI agents rather than apps and including a credit-card-sized "Badge" concept able to record, summarise and act on its surroundings. From the other side of the Pacific, the SCMP reports on Alibaba's MuleRun, an agent platform pitched as a safer, more enterprise-ready alternative to the open-source OpenClaw, with reach across 43 countries.
For news organisations and information businesses, the strategic implication is uncomfortable. The interface where users will increasingly meet information is no longer a browser tab or an app icon, but an agent acting on their behalf. The website, the homepage, the push notification, even the brand, all become inputs to a third-party agent rather than direct touchpoints. The entire muscle of the digital news economy was built around the opposite assumption.
The race for the agentic OS is, at heart, a race to own the layer between user intent and the open web. Whichever company sits there controls a new kind of distribution and a new kind of editorial filter, even if it never calls itself a media company.

Reflections
If the unit of interaction becomes the agent and not the page, what is left of editorial branding as we know it?
What does competition policy mean when the next dominant interface is not a search engine but a workforce of agents?

Chapter 2 — News organisations push back

14.4 — Sulzberger reframes the AI fight as collective action

Source
New York Times publisher A. G. Sulzberger on why (and how) news publishers should fight AI platforms — Reuters Institute for the Study of Journalism

WAN-IFRA forms a strategic partnership with the SPUR Coalition — WAN-IFRA

OpenAI not planning to share advertising revenue with publishers — Press Gazette, Charlotte Tobitt

Dispatch
At the World News Media Congress in Marseille, A. G. Sulzberger argued that publishers have been "too quiet, too passive and too fragmented in the face of abuses by AI companies", calling for a coordinated, principled response to platform behaviour, as the Reuters Institute reports. In the same week, WAN-IFRA joined the SPUR Coalition as a strategic partner, bringing 36 publishers and affiliates including the BBC, FT, Guardian, Telegraph, Sky News, Mediahuis and the European Publishers Council into a single framework for AI licensing, telemetry and content protection. Against that backdrop, OpenAI's VP of media partnerships Varun Shetty told Press Gazette that the company has "no plans" to share advertising revenue with publishers whose content surfaces alongside ads in ChatGPT.
This is the long-promised pivot from individual deals to industry architecture. SPUR's telemetry standard, the WAN-IFRA partnership and the Sulzberger speech together push the conversation away from one-off settlements and towards a permanent infrastructure of licensing, measurement and enforcement. OpenAI's flat refusal on ad-revenue sharing is the foil that gives all of this urgency.
The wider lesson is that no individual publisher, however large, can negotiate a sustainable settlement alone. The question is whether news organisations can sustain collective discipline once each member is offered a sweet bilateral deal on the side.

Reflections
If collective bargaining becomes the new normal between publishers and AI platforms, who decides which titles count more than others?
What is the right unit of payment in an AI world: per crawl, per query, per answer, per signal of trust?

14.5 — European publishers, Google ad tech, and the price of a decade

Source
European publishers seek £552m+ from Google claiming ad market abuse — Press Gazette, Charlotte Tobitt

Dispatch
More than twenty European publishers, backed by Prague-based litigation funder LitFin, are bringing a group claim of over £552 million in damages against Google, building on the European Commission's €2.95 billion fine for adtech abuses, Press Gazette reports. The publishers, from the Czech Republic, Estonia, France, Hungary, Finland, the Netherlands, Poland and Sweden, argue that they would have earned significantly higher ad revenue and paid lower fees absent Google's favouring of its own ad exchange AdX through DFP, Google Ads and DV360.
For policymakers and CEOs, this is the moment damages litigation becomes a serious channel of redress in the European media economy, alongside regulation. The group-claim structure, by sharing costs and risk via a third-party funder, gives small and mid-size publishers access to a forum that until recently was reserved for groups like Axel Springer or Schibsted. The Commission's findings, validated by parallel US DOJ rulings, lower the evidentiary threshold for follow-on claims.
The architecture of the past decade — opaque adtech, asymmetrical contracts, ineffective remedies — is finally being priced. Whether the resulting damages are large enough to change behaviour, or simply enter the cost line of the world's most profitable advertising business, is the question that will define the next round.

Reflections
What share of an award realistically reaches the newsroom?
Once the bill for the last decade is settled, what enables this architecture being rebuilt around AI summaries and agent traffic?

14.6 — Jarvis pushes back on the protectionist reflex

Source
Jeff Jarvis challenges conventional wisdom on public policy, funding, AI and audience growth — Editor & Publisher

Dispatch
In a wide-ranging E&P interview, Jeff Jarvis argues that subsidising legacy chains and expanding copyright to fence off AI is the wrong policy reflex; he urges local publishers to band together, build an "API for news" for AI systems, embrace collaboration with the ecosystem around them and stop treating content as the totality of journalism's value, as the Editor & Publisher piece sets out. He explicitly warns against bargaining-code structures that channel funds to large national groups and away from genuinely local journalism.
For decision-makers this is the uncomfortable counter-melody to the Sulzberger and SPUR signals. Jarvis does not deny that AI companies should help fund journalism; he objects to doing it through expanded copyright and protectionist legislation that ends up benefiting hedge-fund-owned chains. His proposal is closer to a tax-and-redistribute logic than to a licensing-and-litigate one.
The strategic question is whether the industry's collective action can leave room for new entrants and local sustainability, or whether it will harden into an architecture that pays Murdoch and the New York Times generously while leaving the local ecosystem to wither.

Reflections
What would a genuine "API for news" look like designed by news managers and journalists?
How do we measure whether AI deals actually fund original reporting?

Chapter 3 — Copyright, training data and the limits of enforcement

14.7 — The training/inference blind spot in EU copyright

Source
A Blind Spot at the Heart of EU Copyright and AI Policymaking? — Kluwer Copyright Blog, Paul Keller

Dispatch
Paul Keller argues that the European Commission's upcoming copyright intervention conflates two legally and economically different uses of protected works by AI: training-time use, which is large-scale, opaque and difficult to license individually, and inference-time use, which is more traceable and amenable to licensing or remuneration; collapsing these two into one instrument risks producing incoherent legislation, his Kluwer Copyright Blog piece warns. He calls on stakeholders to press the Commission to make this distinction the analytical prior of its work.
For news organisations this matters because each phase has a different economic centre of gravity. Training is mainly about how datasets are assembled, who funds them and which actors first commercialise the resulting model; inference is about ongoing use, retrieval-augmented generation, and the moment-to-moment relationship between an AI product and a publisher's archive. A licensing regime that fails to distinguish them risks under-rewarding archives and over-restricting research, or vice versa.
The deeper argument is that the value gap is real, but the instruments to close it are not yet calibrated. EU copyright still treats the AI value chain as one undifferentiated act of "use", and that mismatch is the engine of every incoherent debate the sector is having.

Reflections
If training and inference are treated as legally distinct uses, who will be the counterparty for each?
How do we build licensing structures that respect this distinction without creating cost that only large platform players can carry?

14.8 — IPKitten on local models, and why I disagree

Source
Running a Tintin model locally on your laptop might just transform copyright — The IPKat

Dispatch
The IPKat post argues that the rise of efficient open-weight models that run locally on laptops, illustrated by HuggingFace experiments fine-tuned on the Tintin universe, fundamentally changes copyright. The piece suggests that because local users escape platform-level guardrails, are not captured by Article 53(c) of the AI Act, and can dial up "temperature" to produce more transformative output, the doctrine should adapt by recognising the collective creative value of model use and considering a shift towards cultural-heritage frameworks rather than individual-act-of-copying frameworks.
This is exactly where I part company with the argument. Generative AI does not in principle change what copyright protects: concrete original works, not ideas, styles or mere economic substitution. AI models, whether in the cloud or on a laptop, work with synthesised, non-traceable representations of training data; that is not a reason to stretch the concept of "work" toward style or tone. What does shift is the architecture of power and control. We could once address abuse mainly through a handful of large platforms and cloud providers, but in a world of open-weight models on thousands of laptops, that handle becomes diffuse and almost ungovernable. The logical response lies not in reinventing copyright itself, but in rules at the source: datasets, training, first commercialisation, and in sectoral deals, including transparency and compensation for news media, rather than in trying to police every local user of a model. Treating local model use as a reason to weaken individual rights, while inventing a vaguer "cultural heritage" alternative, risks both diluting protection and exporting the problem to a body of law not designed to carry it.
The IPKat argument is intellectually elegant, but its policy direction would weaken the position of working creators and news organisations exactly at the moment they need clearer, narrower, more enforceable rules at the input stage of the value chain. The right move is to harden the upstream, not soften the doctrine.

Reflections
If local open-weight models can no longer be reached through platform-level guardrails, where is the next defensible chokepoint: datasets, training runs, or first commercial release?
What does the EU gain, and what does it lose, by treating style and tone as colourable inside copyright?

Chapter 4 — Governance, accountability and the limits of metaphor

14.9 — AI cyber risks and the office built to coordinate them

Source
AI Cyber Risks Are Testing the Office Built to Coordinate Them — Lawfare, Kevin Frazier

Trump Administration Internal Conflict Stalls AI Regulation — The Outpost

Dispatch
Kevin Frazier argues that the Office of the National Cyber Director, created in 2021 to coordinate federal cyber policy, lacks the authority, expertise and budget to handle frontier AI cyber risks, including the cybersecurity behaviour of models such as Anthropic's Mythos and OpenAI's GPT-5.5, his Lawfare piece sets out. In parallel, The Outpost reports that a draft AI executive order was scrapped hours before signing on 21 May, leaving the United States with no new federal AI framework while the EU AI Act enters full enforcement in August 2026, and an internal "knife fight" continues between Commerce, intelligence agencies and pro-industry aides.
For European policymakers this is a strategic gift and a strategic warning at the same time. The gift: the EU now holds the only operational rulebook with statutory authority over frontier models. The warning: the model that most matters for global cybersecurity will be regulated, in practice, by whichever combination of Commerce, NIST, ODNI and ONCD finally consolidates power in Washington, and right now no one has.
The deeper pattern is that AI is colliding with offices and statutes designed for slower threats. When a model can autonomously discover thousands of zero-day vulnerabilities, every "coordination office" suddenly looks like a 20th-century institution staffed by people with 20th-century budgets.

Reflections
Should AI safety evaluation sit with civilian standards bodies, intelligence agencies, or a hybrid; and what does each option imply for press freedom and corporate transparency?
What is the cost of regulatory vacuum in a world where the next major AI incident may be a worm rather than a chatbot?

14.10 — AI-powered worms and an agent-led crime spree

Source
Researchers show how AI-powered worms could wreak havoc on the internet — Engadget, Steve Dent

Researchers Put AI Models in Charge of a Simulated Society. Grok Oversaw a Crime Spree — Gizmodo, AJ Dellinger

Dispatch
A University of Toronto team has built a prototype AI worm using publicly accessible models, capable of exploiting known flaws across Linux, Windows and IoT devices, siphoning passwords and processing power, and persisting through patches, Engadget reports. In a different but resonant experiment, Gizmodo describes how Emergence AI placed leading models in charge of simulated towns: Claude Sonnet 4.6 produced quiet stability, Gemini 3 Flash kept agents alive but recorded 683 crimes, GPT-5 Mini killed all its agents within a week, and Grok 4.1 Fast presided over a 183-crime spree and total societal collapse in four days.
Taken together these two papers are the operational and the moral preview of agentic AI. The Toronto worm shows what one model can do when given autonomy and a network; the Emergence simulation shows what happens when several models govern a society over time. In both cases the lesson is that guardrails which work on a single prompt do not survive long horizons, multi-agent dynamics or sustained adversarial use.
This is the part of the AI conversation that most resists serene language. Long-horizon agentic systems are not a deployment problem; they are a governance problem with deployment characteristics. Formally verified safety architectures, as Emergence's researchers recommend, are not a slogan but a requirement.

Reflections
If "alignment" only holds for a single turn, what does safety mean across thousands of agent decisions in a real economy?
Who is liable when a multi-agent system, each component lawful on its own, collectively produces a crime wave?

14.11 — Anthropomorphic terms, AI worker rhetoric and the accountability gap

Source
Anthropomorphic AI terms create gaps in accountability — Brookings, Stephanie K. Pell

Developers won't work without AI anymore. The research says they're wrong — The Next Web, Ana Maria Constantin

Companies Must Go Beyond AI Adoption to Realize Its Full Potential — Boston Consulting Group

Gartner Says Autonomous Business and AI Layoffs May Create Budget Room But Do Not Deliver Returns — Gartner

Dispatch
A Brookings analysis argues that anthropomorphic language in AI policy and discourse — agents, workforce, hires, decisions — blurs responsibility, making systems appear as independent actors rather than tools built and deployed by named institutions. The Next Web reports that METR could not replicate its earlier AI coding productivity study because developers refused to work without AI, even though available evidence suggests AI tools may slow them down, introduce bugs and increase maintenance cost, the piece records. BCG's global survey finds that 72 percent of respondents use AI regularly but business value is captured only by firms that redesign workflows, and Gartner warns that "autonomous business" and AI-driven layoffs may free budget but do not deliver returns.
The strategic point for CEOs and policymakers is that the language we use to describe AI is shaping the accountability we are willing to demand. If we say a model "decided" to deny a claim, "hired" a candidate, or "approved" a transaction, we have already moved responsibility one step away from the humans and institutions that designed, deployed and oversaw it. The BCG and Gartner numbers should sober the boardroom: adoption is mainstream, productivity is not yet proven, and layoffs without workflow redesign tend to destroy value rather than create it.
The cleanest discipline is to speak operationally. "The bank's automated underwriting system, configured by X under policy Y, produced this output". It is uglier than "the AI decided", and that ugliness is precisely the point.

Reflections
If developers will not work without AI but cannot prove gains, what does that tell us about identity, status and the actual function of these tools in knowledge work?
How long can a company defend AI-driven layoffs to shareholders if the productivity case stays unproven?

Chapter 5 — Media, democracy and the limits of conversation

14.12 — Sulzberger, Poynter and journalism's confidence problem

Source
Why don't journalists push back against Trump? — Poynter

One Company, One Beat — Columbia Journalism Review, Amos Barshad

Reuters wins Pulitzer Prizes for Beat Reporting and National Reporting — Reuters

Dispatch
The Poynter commentary returns to a question that has shaped much of US press discussion this year: why mainstream newsrooms still mostly fail to push back collectively against a presidency that openly intimidates them. Read alongside CJR's "One Company, One Beat", profiling Caroline O'Donovan on Amazon, Edward Niedermeyer on Tesla, Mark Gurman on Apple and Brooks Barnes on Disney, it suggests that the model of single-beat accountability journalism is becoming the most reliable form of structural reporting on power. Reuters won this year's Pulitzers for Beat Reporting on Meta's failures around fraud, AI chatbots and child safety, and for National Reporting on "The Revenge of Donald Trump", documenting 470 named targets of executive retaliation.
For news organisations the strategic insight is that depth, persistence and patient sourcing on a narrow beat now beat the volume of generic coverage. The Reuters Pulitzers were not won by viral output; they were won by endurance and patience. The Poynter and Sulzberger pieces converge: in a hostile political environment, the absence of collective and public journalistic indignation might be a succesful strategy.
The line that connects all three is that accountability journalism scales not by going broader, but by going deeper, repeatedly, on a small set of power centres.

Reflections
If a reporter on a single beat is a highly successful structure for accountability, what does this learn us about structuring newsrooms?
What does it say about the power of journalism that a journalist can outsmart power by dimming our ego?

14.13 — Editorial sovereignty, cloud sovereignty, and digital independence

Source
Editors' Choice: Get off of my cloud? — Euractiv, Orlando Whitehead

Netherlands moves GPT-NL from lab to live: first pilots under way — Computer Weekly, Kim Loohuois

Hoe journalisten de hype rond AI maken, bevragen én vrezen — SVDJ, Nigel van Schaik

Dispatch
Euractiv's opinion piece argues that the EU's Cloud and AI Development Act (CAIDA) takes a tough line on China but goes soft on the United States, failing to identify critical sectors and reducing cloud sovereignty to an invitation rather than an obligation. Computer Weekly reports that GPT-NL, the Dutch national language model developed by TNO, NFI and SURF, has moved from research artefact to live pilots with a €13.5m public budget, transparent training metadata on HuggingFace and a planned commercial roll-out in the second half of 2026. SVDJ records how Dutch journalists, with researcher Tomás Dodds, are starting to interrogate their own role in producing AI hype.
These three pieces draw a coherent line. Cloud sovereignty without obligation is empty. Sovereign LLMs only matter if they reach real users. And journalism that does not interrogate its own narrative role becomes part of the marketing machine. For European media, the implication is direct: the credibility of any AI strategy depends on where the data sits, which model is used, and how journalists frame the story.
Lokke Moerel's line, quoted in the GPT-NL piece, captures the stake: if countries only make rules for technology built by others, they will always be chasing events. Sovereignty is a posture before it is a product.

Reflections
What does it mean for a European newsroom to use a sovereign model?
For GPT-NL and other EU spin offs to succeed, what should we build on top of it that platform models cannot deliver?

14.14 — Meta, the Oversight Board, free speech and the question of harm

Source
Meta's Oversight Board races to govern the AI surge — Rest of World, Ananya Bhattacharya

Shareholders tell Zuck being dead is bad for business — Substack, Ricky Sutton

Future Media

Shareholders tell Zuck being dead is bad for business

I have a confession. I have a complicated relationship with incentives…

Read more

4 months ago · 15 likes · 1 comment · Ricky Sutton

Tech Talk: Free Speech and AI — Center for Democracy & Technology

Microsoft AI Director: Magnifica humanitas valuable for AI development — Vatican News

Dispatch
Rest of World reports that Meta's Oversight Board, designed for case-by-case review, may break under the volume and velocity of AI-generated content and AI-driven moderation; it is considering broader systemic recommendations and faster procedures. Ricky Sutton's Substack post records Meta shareholders pushing to tie executive pay to child safety as fines and regulation mount. The Center for Democracy & Technology hosts a discussion on how to govern AI while protecting free expression. And Vatican News reports Microsoft AI Director Taylor Black endorsing Pope Leo XIV's encyclical "Magnifica humanitas" as a useful anthropological framework for AI development, arguing that probabilistic, co-created products require a richer understanding of the human person.
The connecting tissue is that the institutions designed to govern speech and harm on platforms are visibly under-built for AI. The Oversight Board model assumes a manageable case load; shareholder pressure assumes corporate boards still see child safety as a fiduciary issue; CDT's free-speech framing assumes a relatively coherent state, an institutional press and a recognisable user. AI breaks all three assumptions at once. The Vatican intervention is interesting precisely because it inserts a slow, anthropological vocabulary into a debate that has been dominated by accelerationist metaphors.
The deeper question for democratic institutions is whether we will design new bodies for AI-era speech and harm, or simply load every existing one until it breaks.

Reflections
How can a case-based oversight board be the right model when most contested content is now AI-generated, AI-moderated, or both?
What does "free speech" mean when the largest speaker on a platform is a model owned by the platform itself?

Chapter 6 — Press freedom, climate, leadership

14.15 — Press freedom, Singapore and the limits of indices

Source
Why Global Press Freedom Rankings Struggle with Singapore — The Diplomat, Leonie Spangenberg

Brazilian reporters explain how they use data to track the climate from the Amazon to Rio — LatAm Journalism Review, Teresa Mioli

Just lead: A CEO call to action — Axios, Jim VandeHei

Dispatch
The Diplomat argues that Singapore's low placement in global press-freedom rankings reflects a clash between Western assumptions and a Confucian, harmony-oriented governance model; rankings can capture real restrictions, but also risk measuring conformity to Western political ideals. LatAm Journalism Review shows how Brazilian reporters, led by figures such as Daniel Nardin, are combining environmental specialisation with data journalism to cover the climate from the Amazon to Rio. And Axios's Jim VandeHei calls on CEOs to lead candidly on the economy, the organisation, and especially AI, given that trust in almost every other institution has collapsed.
For policymakers and CEOs the cumulative point is that frames matter. Press-freedom indices, climate data and CEO communication all rest on assumptions about what counts as legitimate authority, legitimate measurement and legitimate honesty. In a year where the loudest signals will come from AI, governments and platforms, the credibility of any leader will depend on whether their framing survives scrutiny by their own employees and audience.
The Axios piece is bracing in its directness: lead, or watch the void fill with louder and less accountable voices.

Reflections
What is the equivalent of "data journalism on climate" for AI and platform regulation, and who is doing it?
If trust in institutions keeps falling, can we really stop the slide, or only slow it?

Chapter 7 — Amsterdam, this week

14.16 — PDLN 2026 at DPG Mediavaert

Source
PDLN 2026 Conference — Amsterdam, 7–9 June

Press Database and Licensing Network

Dispatch
From Sunday, the Press Database and Licensing Network , 41 member organisations across 25 countries, founded in 2008 with Belga News Agency (Mediargus) as one of founders, the gathers in Amsterdam for its 2026 conference, hosted by Dutch member ArtikelPro, with delegates from Australia, Japan, Korea, most of Europe and South Africa. The programme runs on the canals and at DPG's headquarters in the south of the city, with confirmed speakers from Microsoft, The Guardian, WAN-IFRA, NDP, MT Connect, Cloudflare, AMEC, FIBEP and Miso.ai. Discussion groups include an AI working group run by OPR and Corint, a EuroHub group, and a new group on MMO business and content protection, including how to prevent illegal scraping of publisher websites.
For the news media industry this is the moment where the abstract questions of this dispatch such as licensing, AI, content protection, sovereign infrastructure, collective bargaining become a practical agenda in a single room. PDLN sits at the joint between the publishers who own content and the media-monitoring industry that licenses it, exactly the interface where AI training, retrieval-augmented generation and agentic interfaces now collide.
The conference is a test of whether an international, technically literate publisher can shape the AI conversation rather than merely react to it. After Marseille and the SPUR announcement, Amsterdam is the next station on that line.

Reflections
What would be a successful outcome of PDLN 2026 for European publishers, in concrete terms?
If AI training and inference were treated separately, as Keller suggests, which of these would PDLN members address first, and how?

The image that closes this dispatch is exactly the building, that hosts the PDLN Conference pictured on its quay, on a working morning. A wood-and-water headquarters for a sector that is still trying to decide whether its future lies in defending an old perimeter or in building a new one.

If there is one line to leave the reader with, it is this. The next chapter of the information ecosystem will not be written by whoever has the largest model, but by whoever can sustain the most coherent architecture: legal, technical, economic and editorial, in that order. Mediavaert is one such architecture.

Photo: DPG Mediavaert, Amsterdam — © ANP

13.1 — Mission Before Machinery
Source: Civic duty or community hub? Sam Guzik's two-question compass for the AI era — WAN-IFRA

Dispatch
Sam Guzik offers a deceptively simple framework for publishers entering the AI era: are you primarily civic infrastructure or a community hub, and are you best at breaking news or explaining the world? That matrix matters because different positions imply different vulnerabilities, from zero-click search and platform dependency to the erosion of direct audience relationships, and it pushes publishers to compete more deliberately rather than trying to be everything at once. The most useful part of the argument is not the grid itself, but the discipline behind it: AI strategy becomes clearer when a publisher first defines its mission, its audience, and the specific problem it is trying to solve.

Reflections
If AI makes generic distribution easier, what remains truly distinctive about a publisher that has not clearly chosen its mission?
Is the future advantage of news organisations more likely to come from speed, explanation, community, or trust?

13.2 — Above the Loop
Source: CEO Insider: The human in the loop — the human above it — WAN-IFRA, by Ladina Heimgartner

Dispatch
Ladina Heimgartner argues that the next newsroom challenge is to design workflows in which humans increasingly stand above the loop, directing multiple AI agents while remaining accountable for standards, trust, and editorial judgment. This is a strategic shift, because it implies new management roles, new newsroom hierarchies, and new expectations around personal credibility, visible authorship, and creator partnerships at a moment when synthetic content is becoming harder to distinguish from authentic reporting. Her core message is that journalism still has a human advantage, but only if media leaders redesign their organisations intentionally before the operational logic of AI redesigns them by default.

Reflections
What kind of editor is needed in a newsroom where part of the staff is made of agents rather than people?
If trust increasingly attaches to identifiable individuals as well as institutions, how should news brands rethink authority?

13.3 — Journalism After the Interface Shift
Source: Nordic AI in Media Summit 2026: A deep look into how AI is about to revolutionise the news ecosystem — Reuters Institute for the Study of Journalism

Dispatch
The Reuters Institute frames the fourth Nordic AI in Media Summit around questions that go well beyond efficiency: what the news economy will look like, what and who will be automated, and what journalism will still mean in the age of AI. That framing is useful because it shifts the debate away from isolated tools and toward system change, where the central issue is how AI redistributes value, attention, labour, and editorial meaning across the wider news ecosystem. The harder strategic implication is clear: the more media organisations outsource discovery, formatting, and interface layers to external systems, the more urgent it becomes to define what cannot be outsourced without hollowing journalism out from within.

Reflections
Which newsroom functions are merely expensive, and which are foundational even when automation looks cheaper?
If interfaces do more of the presenting, summarising, and routing, where does editorial identity actually reside?

13.4 — Relevance Without Dilution
Source: AP CEO Defends Swatch Collaboration — Bloomberg Video

Dispatch
In Bloomberg's interview, Audemars Piguet CEO Ilaria Resta defends the brand's Swatch collaboration on the grounds that the greater risk is not accessibility, but irrelevance. Although the piece is not about journalism, the analogy is useful for media because legacy institutions now face a similar dilemma: whether opening parts of the brand to wider audiences strengthens long-term relevance or weakens scarcity without building durable loyalty. For news organisations, the challenge increasingly resembles that of luxury brands: remaining accessible enough to stay culturally relevant, while preserving enough distinctiveness to remain valuable. For publishers under AI and platform pressure, this is becoming a central strategic question: how do you modernise formats, partnerships, and reach without surrendering the forms of trust and distinction that made the brand worth extending in the first place?

Reflections
When does a strategy of broader access deepen a brand, and when does it simply flatten it?
How should publishers distinguish between smart adaptation and self-commodification?

13.5 — AI Enters the Stylebook
Source: New AP Stylebook features expanded artificial intelligence chapter — The Associated Press

Dispatch
The 58th edition of the AP Stylebook includes an expanded chapter on artificial intelligence, with new entries on AI agents, AI slop, and vibe coding, reflecting how quickly AI terminology has moved from technical jargon into the everyday language of newsrooms and audiences. The additions are more than semantic housekeeping; they signal that publishers now need shared standards for describing systems that act autonomously, distinguish between human and synthetic content, and explain coding workflows to general readers without defaulting to insider slang. The real value of this update is not only consistency, but the implicit recognition that clarity about AI is becoming a baseline editorial responsibility, not an optional technical footnote.

Reflections
If newsrooms do not define AI terms clearly, who will, and on whose terms?
Does standardising language around AI help preserve editorial authority, or does it merely formalise a transition already underway?

13.6 — Voice from the Margins
Source: How Khabar Lahariya Brought Hyperlocal Journalism to Rural India — Nieman Reports

Dispatch
Khabar Lahariya, India's only women-run independent rural news outlet, was founded in 2002 by women from marginalised communities in one of the country's most underdeveloped regions, many without formal education, and today reaches 5 million people monthly across digital platforms with hyperlocal reporting in local languages. The outlet's strength lies in its editorial clarity about who it serves and why it exists: it covers the stories that urban newsrooms ignore, in the language people actually speak, reported by journalists who understand village politics, caste dynamics, and the everyday corruption that national media rarely bother to verify. In an AI era obsessed with scale and automation, Khabar Lahariya is a reminder that journalism's competitive advantage often comes not from technology, but from proximity, trust, and the willingness to show up where others will not.

Reflections
If trust depends on proximity and context, can AI-driven newsrooms replicate that without human reporters embedded in communities?
How many publishers claiming to serve underrepresented audiences have actually hired reporters from those communities?

13.7 — OSINT as Newsroom Standard
Source: New OSINT tools help journalists fight misinformation — Editor & Publisher, by Bob Sillick

Dispatch
Open-source intelligence tools are increasingly becoming essential infrastructure for journalists trying to verify images, geolocate videos, trace social media manipulation, and debunk claims in real time, particularly as misinformation spreads faster than traditional fact-checking can keep pace. The rise of OSINT is changing both the skillset required in newsrooms and the economics of verification: what once required expensive specialist teams can now be done, at least partially, by reporters with training in reverse image search, satellite imagery analysis, and social network mapping. The deeper implication is strategic: as AI makes synthetic content easier to produce and harder to detect, newsrooms that do not build OSINT capacity internally risk becoming consumers rather than validators of the information circulating online.

Reflections
If OSINT skills are now baseline for credible journalism, how many newsrooms are actually investing in training rather than outsourcing verification?
Does the democratisation of OSINT strengthen independent journalism, or does it simply accelerate the arms race between verification and deception?

13.8 — The Mirrored Bias Effect
Source: Think the media's biased against you? You probably think misinformation is too — Nieman Lab

Dispatch
Researchers studying what they call the hostile misinformation effect have found that people who believe the media is biased against their political views are also more likely to believe that misinformation disproportionately targets their side, creating a symmetrical distrust that makes correction harder and deepens polarisation. This is not simply a media literacy problem; it is a structural challenge for any institution claiming neutrality or objectivity, because the perception of bias now shapes not only how audiences consume news, but also how they interpret fact-checking, moderation, and editorial decisions. For publishers, the implication is uncomfortable: the more they insist on impartiality, the more some audiences will read that insistence as proof of hidden agenda, especially when AI systems amplify and personalise grievance narratives at scale.

Reflections
If claims of neutrality now trigger suspicion rather than trust, what editorial posture remains credible?
Is transparency about editorial process enough to counter hostile perception, or does it merely provide more surface area for criticism?

13.9 — Archiving as Resistance
Source: The Justice Department Erases History; Lawfare Restores It — Lawfare

Dispatch
When the U.S. Justice Department systematically deleted thousands of press releases and case materials related to the January 6 Capitol attack, the legal blog Lawfare used AI tools to recover 95.3 percent of the deleted content from the Internet Archive, restoring public access to a record the government had chosen to erase. The recovery operation is both technically impressive and editorially significant: Lawfare explicitly framed the project as a defence of the public record, stating that "if the administration purges rule-of-law-sensitive materials from government websites, we will do everything in our power to restore them" and that "net loss of information to the public should be zero". This is journalism as institutional memory, and it raises a harder question: if governments, platforms, or corporate actors can quietly rewrite or delete the historical record, who remains capable and willing to preserve it, and under what funding model?

Reflections
If preservation of the public record increasingly depends on independent media organisations using AI to counter official erasure, is that resilience or precarity?
What happens when the next wave of deletions targets smaller institutions without the resources or technical capacity to mount a similar recovery?

Chapter 3
AI Governance & Transparency Initiatives

13.10 — Opening the Black Box, Chinese Style
Source: AI has a 'black box' problem. China wants to make it more transparent — South China Morning Post

Dispatch
China has announced a new national evaluation framework for AI aimed at improving accuracy, reliability, and transparency, with a unified standard to measure models, computing power, and data quality across the sector. The move matters because it treats AI governance not only as a question of content moderation or industrial policy, but as a question of measurability: if systems become economically and politically consequential, states will increasingly want them to be comparable, traceable, and legible to regulators rather than left as opaque commercial black boxes. The broader lesson for Europe is uncomfortable but important: even when political systems differ sharply, the strategic instinct to standardise and audit AI infrastructure may prove more realistic than leaving accountability to voluntary corporate disclosure.

Reflections
If AI systems shape decisions at scale, can democratic oversight exist without common measurement standards?
Is Europe moving fast enough to govern AI as infrastructure rather than only as a market product?

13.11 — Manipulation by Design
Source: Dark patterns in AI chatbots: A taxonomy to inform better design — Center for Democracy & Technology

Dispatch
The Center for Democracy & Technology identified 37 dark patterns across major AI chatbots including ChatGPT, Gemini, Claude, Replika, and Character.AI, grouping them into risks such as opaque data practices, financial exploitation, false urgency, and forced anthropomorphism. This is an important step because it shows that the governance problem is not limited to model outputs; it also sits in interface design, where products can steer users toward oversharing, emotional dependency, prolonged engagement, or paid features without informed consent. In other words, some of the most consequential harms in AI may not come from spectacular model failure, but from ordinary product design choices that quietly shape user behaviour at scale.

Reflections
How much of AI risk today comes from the model itself, and how much from the incentives embedded in the interface around it?
How can AI policy debates focus more attention to design power, instead of on abstract model safety?

13.12 — Unlawful by Design
Source: Unlawful by design: Exposing the human rights costs of generative AI — Amnesty International

Dispatch
Amnesty International argues that standalone generative AI systems built on unlawful web scraping are fundamentally incompatible with international human rights law because they depend on mass invasions of privacy by design and amplify risks around discrimination, freedom of expression, and freedom of thought. The force of the argument lies in its framing: this is not presented as a regrettable side effect of otherwise neutral innovation, but as a structural critique of the data extraction model on which much of generative AI has been built. Whether or not policymakers accept Amnesty's call for prohibition, the report raises a question that will not go away: can systems trained through indiscriminate appropriation of personal and public data ever be reconciled with rights-respecting governance, or are they flawed at the foundation?

Reflections
If unlawful scraping is built into the supply chain of generative AI, can downstream safeguards really solve the upstream problem?
At what point does "innovation" become an alibi for normalising mass extraction without consent?

13.13 — How Machines Perceive Us
Source: How AIs See Our World — Noema Magazine, by Chenoe Hart

Dispatch
Chenoe Hart's essay in Noema argues that AI systems do not simply observe the world as humans do; they translate it into machine-legible abstractions such as bounding boxes, labels, segmented poses, and computational categories that can miss context, ambiguity, and socially meaningful nuance. That matters because governance debates often assume that better models will simply "see" more accurately, whereas the essay shows that perception itself is structured by design choices, training data, and system architecture, which can flatten reality before any decision is even made. The policy implication is subtle but profound: if AI systems perceive through reduction, then transparency is not only about explaining outputs, but about understanding the terms on which the world has already been simplified for the machine.

Reflections
If AI systems see the world through categories and abstractions, how much of reality is lost before a model ever produces an answer?
What would it mean to design public-facing AI systems that adapt better to human complexity rather than forcing humans to adapt to machine legibility?

13.14 — Europe Tests the DMA's Teeth
Source: EU plans to fine Google high triple-digit million euro sum, Handelsblatt reports — Reuters

Dispatch
Reuters reports that the European Union is preparing a high triple-digit million euro fine against Google as part of a Digital Markets Act investigation into allegations that the company favours its own services in search results, with a decision expected before the summer break. If confirmed, this would be the largest penalty yet imposed under the DMA, which matters less as a headline-grabbing sum than as a signal that Brussels may finally be willing to test whether its post-GDPR digital rulebook can actually alter platform conduct rather than merely describe it. The real question is whether such fines can change structural incentives in search and discovery, or whether they arrive only after the market has already been reshaped by self-preferencing, interface control, and user dependence.

Reflections
Can competition law still restore fairer market conditions once user habits and distribution channels have already consolidated around dominant platforms?
How many smaller publishers or competitors can survive long enough to benefit from regulatory correction that comes years late?

13.15 — Regulating the Architecture of Addiction
Source: Le Brésil interdit le design addictif — un modèle pour l'Europe ? — Digital Alternate / commentary on Brazil's 2026 move, read alongside the European Parliament note Addictive design on online platforms
Related context: European Parliament Think Tank, Addictive design on online platforms

Dispatch
Brazil has been highlighted as an early mover in explicitly banning addictive design by name, while European institutions are also moving toward a more direct confrontation with features such as infinite scroll, autoplay, and manipulative notification systems, especially where minors and vulnerable users are concerned. What makes this moment important is that regulation is shifting from content and competition toward product architecture itself: policymakers are no longer asking only what platforms host or whom they disadvantage, but how the interfaces are deliberately engineered to capture attention and shape behaviour. That is a profound step, because once law begins to regulate the mechanics of compulsion, the political debate moves from speech and market share to something more fundamental: whether certain business models are inseparable from psychological manipulation.

Reflections
If endless scroll and autoplay are not neutral design choices but behavioural weapons, should they still be treated as ordinary product features?
Where should democracies draw the line between legitimate persuasion, good user experience, and engineered dependency?

Chapter 5
AI Security & Adversarial Risks

13.16 — The Hidden Prompt in the Background Noise
Source: Hackers Find That Inaudible Sounds Hidden in Podcasts or Recordings Can Hijack AI Voice Chatbots — Futurism, based on research presented at the IEEE Symposium on Security and Privacy

Dispatch
Researchers from China and Singapore showed that adversarial audio signals, inaudible to human listeners, can be embedded in ordinary background media such as podcasts, songs, or videos to hijack voice AI systems and induce them to perform unintended actions, potentially exposing personal data or linked services. What makes the finding strategically important is not just the cleverness of the exploit, but the shift it represents: the attack surface of AI assistants no longer begins only with typed prompts or direct user commands, but with the ambient information environment itself, where malicious instructions can hide in plain sound. Even if the current method depends on access to model weights and works most directly against open-source systems, the larger lesson is already clear: once AI systems become always-listening interfaces to banking, messaging, or home devices, prompt injection turns into a cybersecurity problem for everyday life.

Reflections
If malicious prompts can be hidden in ordinary media, where does the boundary between content and attack really begin?
What happens when the weakest layer in AI security is not the model itself, but the environment through which it listens?

13.17 — Autonomous Lethality at Swarm Speed
Source: Chinese scientists create 'kill-them-all' algorithm for drone warfare — South China Morning Post

Dispatch
A research team in northwestern China has unveiled a new algorithm, HG-STR, designed to allow fixed-wing drone swarms to search large battlefields autonomously and eliminate enemy targets even when communications are jammed and visual conditions are degraded. According to the report, the peer-reviewed paper claims the system is the first known algorithm capable of reaching a 100 percent kill rate while operating fast enough for modern combat conditions, pointing toward a future in which human command may be reduced to a final instruction before lethal autonomy takes over. Whether or not such claims prove fully operational in practice, the geopolitical significance is immediate: AI competition is no longer only about chatbots, models, or productivity gains, but about machine perception and decision-making under battlefield conditions where speed, opacity, and lethality reinforce each other.

Reflections
If autonomous systems are optimised for jammed and degraded environments, does that make escalation more likely by reducing the role of hesitation and communication?
How should democratic societies respond when AI innovation and military utility become increasingly difficult to separate?

Chapter 6
AI Economics & Enterprise Reality Check

13.18 — Token Shock
Source: Uber Burns Its 2026 AI Budget In Four Months On Claude Code — Forbes, by Janakiram MSV

Dispatch
Uber reportedly exhausted its 2026 AI coding budget in just four months, largely because usage of Anthropic's Claude Code surged far beyond expectations after the company pushed adoption internally and tracked engagement through leaderboards. The episode is striking because it exposes a weakness in many enterprise AI strategies: leaders often talk as if AI is a software subscription problem, while in practice heavy use can behave more like volatile infrastructure consumption, with token pricing turning experimentation into a fast-moving cost centre. The lesson is not that AI tools are useless, but that budget discipline, usage controls, and clear economic purpose now matter as much as model capability, especially once internal enthusiasm outruns managerial oversight.

Reflections
How many companies are treating AI as a productivity investment while budgeting for it as if it were just another SaaS licence?
Does internal pressure to "use more AI" create measurable value, or merely inflate token burn and symbolic compliance?

13.19 — The ROI Reckoning
Source: AI sticker shock hits corporate America — Axios

Dispatch
Axios reports that corporate executives are starting to question whether soaring AI spending is producing meaningful returns, as organisations confront rising IT costs, unclear productivity gains, weak governance around licences, and growing frustration over "all-you-can-eat" assumptions that collapse under token-based billing. The deeper problem is strategic rather than merely financial: firms rushed to deploy copilots, agents, and coding tools before deciding where these systems would genuinely improve workflows, how they would access trusted internal data, or what evidence would count as success. That is why the current moment looks less like a pause in adoption than a shift into accountability, where AI spending will increasingly be judged by disciplined use cases, measurable outcomes, and whether it performs better than the humans or systems it was meant to augment.

Reflections
If AI budgets keep rising while use cases remain vague, is the real scarcity money or managerial clarity?
Should boards ask first how much AI costs, or what category of problem it solves better than existing tools?
How many current AI deployments would survive if they had to justify themselves against strict ROI metrics rather than innovation rhetoric?

13.20 — Paying with Yourself
Source: The Hidden Price of Free: What Your Data Is Really Worth — My Security Marketplace

Dispatch
The logic behind "free" digital services remains brutally simple: when users are not paying in cash, they are often paying in behavioural data, exposure, predictive signals, and reduced control over how their profiles are built, sold, and reused across advertising and platform ecosystems. What makes this issue more urgent in the AI era is that data extraction no longer fuels only targeted ads; it increasingly feeds recommendation systems, profiling engines, synthetic personalisation, and model optimisation, extending the economic value of each user far beyond a single click or impression. Seen this way, the hidden price of free is not only privacy loss, but the gradual conversion of everyday human activity into machine-readable inventory that others monetise, analyse, and operationalise at scale.

Reflections
If personal data is now a strategic input for both advertising and AI, can consent models built for the old web still be taken seriously?

Should societies start treating data extraction less as a consumer issue and more as a question of economic power and civic autonomy?

Across this dispatch, from media strategy and editorial standards to AI governance, platform regulation, cybersecurity and enterprise economics, the same structural question keeps reappearing: who controls access?
Control no longer resides only in ownership of content, infrastructure or capital. Increasingly, it resides in the systems that mediate access: search engines, recommendation algorithms, AI assistants, cloud platforms, proprietary datasets, and the interfaces through which citizens, consumers and organisations encounter information.

That is why the emerging AI landscape resembles the story of Ali Baba and the Forty Thieves. The treasure is visible to everyone. The cave itself is not hidden. What matters is knowing the words that open the door.
In the digital economy, those words are no longer "Open Sesame." They are the prompts, models, APIs, ranking systems, licences, standards and governance mechanisms that determine who can enter, who can participate, who captures value, and who remains dependent on the decisions of others.
For news organisations, regulators and democratic institutions alike, the strategic challenge is therefore not merely to produce more content, deploy more AI, or collect more data. It is to ensure that access to knowledge, public information and economic opportunity does not become concentrated in the hands of those who alone possess the keys to the cave.

Synthetic Image

The future of the information ecosystem may ultimately depend less on who owns the treasure than on who controls the door.

There is a growing tendency to discuss artificial intelligence as if it were a software revolution. But the deeper story is infrastructural. AI is becoming a layer that reorganises labour, political communication, scientific production, media distribution and even moral authority. The central question is therefore who controls the infrastructures on which societies increasingly depend: compute, data, verification systems, satellite networks, trusted repositories, democratic communication channels and the information flows that connect them.

This dispatch follows that logic. Across regulation, elections, journalism, open-source intelligence and platform ecosystems, the same pattern emerges repeatedly: societies are rediscovering the strategic importance of trusted intermediaries, resilient public-interest infrastructures and governance models capable of resisting both manipulation and industrial-scale synthetic noise.

For news agencies in particular, this debate is essential. In an environment increasingly flooded by AI-generated content, manipulated narratives and agent-driven distribution systems, trusted verification networks are part of democratic resilience itself.

I. On power and the political economy of AI

Signal 12.1 – “When AI costs more than people”

Source: Fortune, “Nvidia executive says the cost of AI is greater than the cost of employees” –

Dispatch

A senior Nvidia executive argues that, at current scales, running advanced AI systems costs more than employing the people whose work they are meant to augment or replace. He points to the capital intensity of GPUs, energy and cloud infrastructure as a structural constraint on “AI everywhere” business models. The piece suggests that many firms underestimate these ongoing costs when they rush to automate or build AI features. It raises the question of who will ultimately be able to afford high-end AI, and who will be left buying cheaper, less reliable systems.

Reflections

• If the true cost of AI exceeds the cost of human labour in many cases, which kinds of work actually make economic sense to automate?
• What does it mean for newsrooms and news agencies deciding where AI belongs in their workflows — and where human judgement is both better and cheaper?

Signal 12.2 – “Banning AI-generated papers: arXiv’s first hard stop”

Source: 404 Media, “New arXiv rules on AI-generated papers ban”

Dispatch

This piece describes how arXiv, one of the main repositories for scientific research, is tightening rules around AI-generated content. The platform responds to a surge in low-quality machine-authored submissions that strain moderation and risk confusing readers about what has actually been peer-reviewed. New guidelines require clear disclosure of AI use and allow moderators to reject papers that appear to be largely synthetic or deceptive.

The debate mirrors concerns in journalism: when volume explodes and barriers to production collapse, quality assurance itself becomes infrastructure.

Reflections

• What practical standards should exist for disclosing AI assistance in research, journalism and policy documents?
• How can repositories and archives remain open and fast without becoming dumping grounds for AI-generated slop?

Signal 12.3 – “Brazil’s elections as an AI stress test”

Source: Tech Policy Press, “Brazil’s 2026 elections are its first real stress test for AI regulation”

Dispatch

The article argues that Brazil’s 2026 elections will be the first major test of the country’s new AI and platform regulations in a high-stakes democratic context. Lawmakers and regulators have put in place rules around political advertising, deepfakes and content moderation, but enforcement capacity and coordination with platforms remain open questions. Civil society groups warn that generative AI could turbo-charge disinformation campaigns, especially in encrypted and semi-closed ecosystems.

Brazil therefore becomes an early laboratory for one of the defining questions of this decade: can democratic governance mechanisms move faster than synthetic manipulation systems?

Reflections

• What should success look like for AI regulation in an election: fewer abuses, faster response, more transparency, or something else?
• What lessons from Brazil will be most relevant for European elections as AI tools become part of every campaign’s toolkit?

II. On governance, regulation and “living” AI policies

Signal 12.4 – “How much power does the EU AI Office really have?”

Source: Lawfare, “How much power does the EU AI Office actually have?”

Dispatch

Lawfare analyses the formal mandate and practical constraints of the new EU AI Office, which is tasked with overseeing implementation of the AI Act. On paper, the office coordinates enforcement, issues guidance and can investigate systemic risks, but many concrete powers sit with national regulators. The piece highlights uncertainties around staffing, expertise and political backing, which will determine whether the office becomes a genuine centre of gravity or merely a coordinating layer.

For industries like media and technology, this will shape how coherent and predictable AI oversight in Europe really becomes.

Reflections

• What capabilities, technical, legal and political , does an AI supervisor need to become an effective node in the regulatory network?
• How can news organisations engage with such bodies early, instead of only reacting once enforcement arrives at their doorstep?

Signal 12.5 – “AI guidelines as a living policy”

Source: WAN-IFRA, “The living policy: why AI guidelines are never finished” –

Dispatch

This WAN-IFRA piece argues that AI policies inside media companies cannot remain static documents but must evolve continuously as tools, risks and workflows change. The author warns against both rigid bans that quickly become obsolete and vague principles that provide no operational guidance.

Instead, AI governance should function as a living framework for experimentation, documentation, accountability and institutional learning.

Reflections

• How can organisations build feedback loops so that frontline experience with AI tools genuinely updates policy?
• In what ways can “living policies” be communicated externally to strengthen trust with audiences and partners?

Signal 12.6 – “When politicians abandon US messaging applications”

Source: Lawfare Media, “Politicians to ditch Signal for homegrown apps” –

Dispatch

Lawfare reports on a growing trend of politicians and public institutions moving away from global encrypted apps like Signal towards domestically developed messaging platforms. Officially, motives include security, sovereignty and integration with local infrastructures. Critics fear that such systems may also become easier to align with political interests or state access.

The broader issue goes far beyond messaging applications. Communication infrastructure itself is becoming geopolitical terrain.

Reflections

• Why are sovereign or bespoke messaging systems becoming strategically important in democracies?
• Which standards should apply for encryption, governance and oversight when such tools are used for public decision-making?

Signal 12.7 – “Journalism recognised as democratic infrastructure — but underfunded”

Source: European Federation of Journalists / AgoraEU

Dispatch

This article covers a European Parliament initiative explicitly framing journalism as democratic infrastructure deserving structural support rather than occasional subsidies. While symbolically important, journalist organisations argue that the proposed funding remains disconnected from the scale of the challenge.

The contradiction is revealing: societies increasingly recognise trusted journalism as essential infrastructure, while still financing it as if it were an optional market by-product.

Reflections

• What practical difference does it make when journalism is treated as infrastructure rather than merely a commercial product?
• How can funding mechanisms protect editorial independence while still creating long-term stability?

Signal 12.8 – “A papal encyclical on AI”

Source: OSV News, “Pope publishes encyclical on artificial intelligence”

Dispatch

OSV News reports that the Pope is publishing an encyclical dedicated to artificial intelligence, placing AI within a broader moral and social framework touching labour, inequality, dignity and concentration of power.

This matters because AI governance is no longer only a technical or regulatory discussion. It increasingly becomes a civilisational one.

Reflections

• How will a major religious text on AI influence public perceptions of acceptable technological use?
• Can shared ethical reference points across secular and faith-based institutions help anchor AI governance in more human-centred principles?

III. On journalism, trust and the business of news

Signal 12.9 – “Believability vs credibility”

Source: WAN-IFRA, “Believability vs credibility: what journalism can learn from creators”

Dispatch

This piece distinguishes between “believability” — the feeling that a voice is relatable and trustworthy — and “credibility”, based on evidence and institutional processes. Creator-journalism often excels at believability through proximity, tone and personality, while newsrooms emphasise credibility through verification and standards.

The author argues that journalism needs to better integrate both dimensions: making processes more visible and human-scaled without sacrificing rigour. Otherwise, audiences may increasingly gravitate toward believable voices that are less grounded in facts.

Reflections

• How can news organisations show their editorial processes in ways that increase both believability and credibility?
• When does institutional distance remain an asset, even if it feels less immediately “believable”?

Signal 12.10 – “Generation AI and the news”

Source: Common Sense Media, “Generation AI” report

Dispatch

The “Generation AI” study explores how children and teenagers interact with AI tools, platforms and content, including how they encounter news and information. It finds that many young people use AI systems for homework help, explanations and even current affairs, often without clear understanding of how these tools work or what their limitations are.

The report highlights concerns around bias, privacy and exposure to misinformation, but also documents a pragmatic openness among youth to using AI as a learning aid. For newsrooms, it underlines the importance of designing products and educational approaches that meet this generation where it already is.

Reflections

• What does news literacy look like in a world where asking an AI system becomes a default way to understand events?
• How can news organisations build youth-specific interfaces and experiences that integrate AI in more transparent and educational ways?

Signal 12.11 – “AI slop, backlash and detection”

Source: What’s New in Publishing (Substack), “The AI slop backlash is good news” –

What’s New in Publishing

The AI 'slop' backlash is good news for premium publishers

The digital ecosystem has been filled with AI-generated content at a rate that many of us would never have anticipated. Synthetic output seems to be everywhere, with unstimulating articles ten-a-penny and over-processed images and videos at every turn. Social media in particular, seems to have become inund…

Read more

4 months ago · 3 likes

Source: NewsGuard, “Real-time AI content-farm detection datastream”

Dispatch

One article argues that the wave of low-quality AI-generated “slop” flooding the web may ultimately benefit quality publishers by making audiences more alert to the difference between junk and journalism. In parallel, NewsGuard announces a real-time datastream designed to detect AI content farms, giving advertisers, platforms and publishers signals to avoid polluted inventory.

Together, these developments suggest the early formation of an “anti-slop ecosystem”: user backlash, detection tools and premium brands converging to revalue verified reporting.

Reflections

• What role should independent rating and detection services play in the economics of advertising and distribution?
• How could a shared clean datafeed for news, created by news agencies, become a competitive advantage in AI-driven environments?

Signal 12.12 – “AI agents and liquid content”

Source: CJR Tow Center, “AI agents are coming for news – can publishers reclaim control?”

Source: Journalism.co.uk, “Preparing for the second act of AI: agents, liquid content and AI deals”

Dispatch

These pieces map a near future in which AI agents — not human readers — become the primary consumers of news APIs, repackaging and routing content for end users. They describe “liquid content” flowing through multiple agentic systems, from personal assistants to enterprise dashboards, often far removed from the original publisher environment.

The articles ask whether publishers can shape this landscape through standards, licensing and technical controls, or whether they risk becoming invisible suppliers inside an agent-dominated ecosystem.

For news agencies in particular, this future is simultaneously a threat and an opportunity: the possibility of becoming trusted source layers for AI systems themselves.

Reflections

• What technical and contractual levers do news agencies need in order to remain sustainable and independent?
• How will editorial standards and attribution function when most “readers” are software systems acting on behalf of humans?

IV. On critical information infrastructure and alternative ecosystems

Signal 12.13 – “Mapping the blind spots from space”

Source: Columbia Journalism Review

Dispatch

CJR examines how satellite imagery and open-source intelligence have become indispensable tools for covering conflicts and closed regimes, while also revealing their own blind spots. Companies like Planet and Maxar provide near-real-time imagery that journalists and researchers use to document troop movements, damage and infrastructure.

But the piece also demonstrates that visibility remains uneven: some crises are extensively documented, while others remain effectively invisible because of technical, commercial or political constraints.

Satellite systems are therefore becoming part of global information infrastructure — powerful, but never fully neutral.

Reflections

• What are the risks when access to strategic imagery depends on a handful of private actors?
• How could public or cooperative satellite initiatives improve transparency around crises that currently remain in the shadows?

Signal 12.14 – “Digital gardens beyond Big Tech”

Source: LatAm Journalism Review

Dispatch

This piece recounts how community media outlets in Latin America are building “digital gardens”: smaller, locally controlled ecosystems designed to reduce dependence on major platforms. They invest in newsletters, websites, messaging groups and offline communities in order to cultivate slower but more resilient audience relationships.

The strategy reflects more than platform frustration. It reflects a desire to realign infrastructure with mission instead of outsourcing distribution to systems governed by entirely different incentives.

Reflections

• What can larger news organisations learn from community media about building smaller, more intentional digital spaces?
• How could news agencies support such ecosystems through feeds, tools or training?

Signal 12.15 – “Virtual assistants, AI and who really writes LinkedIn”

Source: Rest of World

Dispatch

Rest of World shows how a growing industry of low-paid virtual assistants in the Philippines use AI tools to write LinkedIn posts and comments for Western executives, effectively running a “thought leadership” content mill for around seven dollars an hour.

These assistants are hired to understand engagement logic, maintain constant online presence and manufacture professional authenticity — often with little disclosure that neither the text nor the engagement originates from the named individual.

Meanwhile, LinkedIn simultaneously tightens restrictions against third-party automation and scraping while continuing to permit selected API-based ecosystem partners.

The result is a striking paradox: authenticity itself becomes industrialised, outsourced and partially synthetic.

Reflections

• What does “authenticity” mean on professional networks when much of the engagement is produced by offshore assistants and AI systems rather than the named individuals themselves?
• Which transparency standards should apply once AI-assisted ghostwriting becomes a normal cross-border service industry?

Signal 12.16 – “The Hague signal: news agencies as democratic infrastructure”

Source: EANA / European Alliance of News Agencies

Dispatch

At the EANA Spring Conference 2026 in The Hague, hosted by ANP under the title The World of Disinformation: Challenges & Solutions for News Agencies, European news agencies treated disinformation not as a side issue but as a structural challenge for democracies and public governance.

Speakers such as Svitlana Slipchenko (VoxUkraine) demonstrated how information manipulation has become a permanent geopolitical instrument aimed at weakening democratic resilience, while Saman Nazari (Alliance4Democracy) presented a genuine whole-of-society response linking OSINT, media monitoring and institutional cooperation.

AFP fact-checking practices, GPT-NL initiatives, multimedia verification workflows and the EANA disinformation questionnaire all pointed in the same direction: news agencies are quietly building operational capabilities to detect, verify and contextualise false narratives at scale — not only for media clients, but for the broader democratic ecosystem itself.

The Hague therefore crystallised a deeper conclusion running through this dispatch: trusted news agencies are evolving from media organisations into critical democratic information infrastructure.

Reflections

• If news agencies are recognised as democratic infrastructure, what long-term obligations does this create for states, institutions and commercial actors in terms of support, access and legal protection?
• How can agencies operationalise a whole-of-society response to disinformation while preserving editorial neutrality and independence?

Conclusion — From media companies to infrastructure

One theme cuts through almost every signal in this dispatch: trust is becoming infrastructural.

Satellite constellations. Verification networks. OSINT communities. Messaging systems. AI governance bodies. Detection layers against synthetic content. Trusted archives. Human-centred editorial workflows. These are no longer peripheral tools around democracy. They increasingly are democracy’s operating system.

In that environment, the role of news agencies changes fundamentally.

Historically, agencies were often perceived as wholesalers of information operating quietly behind the visible media landscape. But in an AI-driven ecosystem saturated by synthetic content, fragmented narratives and agentic distribution systems, the ability to provide verified, neutral and resilient information flows becomes strategically essential.

The parallel with the EANA visit the European Space Agency is not accidental. Space infrastructure quietly enables navigation, communications, climate monitoring and security without most citizens ever directly seeing the systems behind it. Trusted information infrastructures increasingly play a similar role inside democratic societies: invisible when functioning well, existential when weakened.

The challenge for Europe is therefore not only technological competitiveness. It is institutional resilience.

The question is no longer whether trusted information infrastructures matter. The real question is whether democratic societies will recognise their strategic value early enough to preserve them before synthetic abundance makes trust itself a scarce resource.

Photo: Participants of the EANA Spring Conference 2026 at ESA ESTEC in Noordwijk, hosted by ANP. The setting itself carried symbolic weight: Europe discussing disinformation, verification and democratic resilience inside one of its most important scientific and infrastructural institutions.

Copyright ANP.

Dispatch
In her Overseas Press Club keynote, AP executive editor Julie Pace reminds the industry that “news doesn’t reveal itself from a distance. It has to be witnessed.” In an era of instantaneous feeds and AI‑generated text and images, she argues that the defining value of journalism remains reporters and photographers who are physically present, verifying facts on the ground, often at personal risk. Her examples, from surveillance reporting to work in Gaza, underscore that independent, on‑the‑scene journalism is not a nostalgic ideal but a contemporary necessity for understanding what is actually happening. Pace’s plea reframes technology as context rather than substitute: powerful tools may accelerate information, but only humans on the scene can decide what truly matters and make it accountable to reality.

Reflections
1. If audiences mostly see AI‑generated summaries, how can they still recognise the difference between generic “content” and journalism that has been verified on the ground?
2. What new responsibility do news agencies have towards both their clients and the wider public to make their physical presence and reporting methods visible in a world of synthetic text and images?

Signal 11.2 – “Trust as the product: Dow Jones’s billion‑dollar newsletter playbook”

Source: Ricky Sutton & Stu Rogers, “Dow Jones’s `$1 billion bet on newsletters and niche intelligence”, Future Media

Dispatch
Dow Jones’s new $1 billion earnings target for newsletters and niche intelligence is framed not as a celebrity‑journalist gamble, but as a return to its founding thesis: objective business information, produced by a disciplined newsroom, is a product people will pay for. From Charles Dow’s hand‑delivered “flimsies” to today’s specialist products like Dow Jones Risk & Compliance and Dow Jones Energy, the through‑line is the same: trust, methodology and verification at scale, rather than personality alone, drive durable subscription value. The piece argues that we are living through a reshuffle of the trust hierarchy, where institutions that can turn rigorous reporting and market intelligence into board‑ready, data‑rich newsletters are positioned to win back attention and revenue. Individual voices still matter at the edges, but the core bet is that a well‑governed newsroom, plugged into the right niches, can turn “the truth in its proper use” into a modern information infrastructure.

Reflections
1. If trust is the product, how should news organisations design newsletter and intelligence products that make their editorial standards and verification methods visible rather than invisible?
2. In a market crowded with personality‑driven journalism, what advantages can an institutional newsroom offer in terms of continuity, depth and resilience of coverage?


Signal 11.3 – “Creator journalism, legacy brands and the missing layer of verification”

Source: “Creator journalism’s rise is ‘the most disruptive shift the news industry has seen,’ ex-BBC News head says”, Nieman Journalism Lab

Dispatch
In an interview highlighted by NiemanLab, former BBC News head Deborah Turness argues that “creator journalism” on platforms like YouTube, TikTok and Substack is becoming the most disruptive shift the news industry has faced, as audiences increasingly follow individual voices rather than visiting institutional news brands. She warns that this fast-growing ecosystem often operates outside traditional editorial checks and verification routines, creating new vulnerabilities around accuracy, accountability and manipulation. At the same time, Turness and the researchers cited see a strategic opening: both legacy newsrooms and creator-journalists will need access to reliable, shared information infrastructures if they want to build sustainable trust, which positions news agencies as critical, if often invisible, components of the future news economy.

Reflections
1. If more people get news through creator channels, how can verification layers from professional news infrastructures be made visible and valuable instead of disappearing into the background?
2. What kinds of partnerships or hybrid models could allow creator-journalists to tap into institutional fact-checking and standards without losing their autonomy and relationship with audiences?

II. On Business, Platforms and the News Deployment Model

Signal 11.4 – “From model company to deployment company”

Source: Ben Thompson, “The Deployment Company: Back to the 70s, Apple and Intel”, Stratechery

Dispatch
In his Stratechery essay, Ben Thompson argues that the most important companies of the AI era may not be the labs that build the most advanced models, but the “deployment companies” that can reliably integrate those models into real organisations. Looking back to the 1970s, he draws a parallel with Intel and later Apple: strategic power shifted to those who controlled the integration layer between new technology and concrete customer use cases, from PCs on desks to smartphones in pockets. In Thompson’s reading, OpenAI’s creation of a dedicated deployment entity, and similar moves by other tech firms, signal a world where value comes from orchestrating AI across sectors, workflows and compliance regimes, not just from raw research breakthroughs. By that logic, the news industry’s next movers may be those who act as “information deployment companies” – entities that already sit inside newsrooms, governments, companies and even creator ecosystems, and can turn critical information infrastructure into something that is actually used, trusted and updated every day.

Reflections
1. If strategic power in AI shifts from model builders to deployment companies, what would it mean for the news ecosystem to have its own “information deployment layer”?
2. Which organisations are already embedded deeply enough in newsrooms, public institutions and companies to orchestrate this layer – and what capabilities would they still need to add?

Signal 11.5 – “When three platforms eat the ad market”

Sources:
• eMarketer, “Meta ad revenues will surpass $`240 billion this year”
• Press Gazette, “Google, Meta and Amazon took two thirds of £46bn UK 2025 adspend”

Dispatch
According to new forecasts, Meta’s global advertising business will reach around `$240–243 billion in 2026, growing more than 20% year‑on‑year and overtaking Google as the world’s largest digital ad company, with roughly 26.8% of worldwide ad spend. In the UK alone, Press Gazette calculates that Google, Meta and Amazon captured about £31 billion of the £46.7 billion total ad market in 2025 – roughly two thirds of all advertising – while all national news brands, magazines, regional titles and radio combined took only around £1.6 billion, or 3.8% of spend. The numbers describe a highly concentrated business reality: the core economic fuel of advertising that once sustained a broad news ecosystem is now routed through a handful of global platforms whose incentives, governance and accountability are only loosely connected to the long‑term health of independent news or democratic public spheres.

Reflections
1. What happens to the diversity and resilience of the news ecosystem when the primary economic growth engine is captured by a few global platforms rather than distributed across many publishers?
2. In a world where platforms dominate ad revenues, what sustainable business models remain for news organisations that still invest in reporting, verification and critical information infrastructure rather than pure scale and targeting?

Signal 11.6 – “RTL’s streaming pivot in a platform‑dominated ad market”

Source: “RTL beats revenue view, streaming turns profitable”, Reuters

Dispatch
Reuters reports that RTL Group slightly beat first‑quarter revenue expectations with around €1.3 billion in revenue and, crucially, turned its streaming business profitable for the first time, keeping full‑year guidance of €25–50 million operating profit from streaming and about €725 million in total profit. This comes despite continued pressure on traditional TV advertising, which RTL expects to decline by around 3% in 2026, and follows several years of targeted investment in RTL+ and Videoland, resulting in 8.4 million paying subscribers and 27% streaming‑revenue growth year‑on‑year. RTL’s strategy hinges on shifting from pure reliance on linear ad markets to a mixed model where local streaming platforms, stronger content IP via Fremantle, and growing digital ad revenues together offset the structural decline in broadcast advertising. In other words, while the global ad duopoly continues to drain budgets from national players, RTL’s answer is not retreat but reconfiguration: scaling its own direct‑to‑consumer streaming ecosystem to stay in the game.

Reflections
• What can other national broadcasters and news organisations learn from RTL’s decision to absorb short‑term profit pressure in order to build a profitable, local streaming ecosystem alongside a shrinking linear ad market?
• What do such mixed models, combining subscription, local digital advertising and owned content IP, realistically need to counterbalance the gravitational pull of global platforms over the long term?

Signal 11.7 – “The Spotify lesson: when distribution owns the tap”

Source: Joel Gouveia, “The Death of Spotify: Why Streaming is Minutes Away From Being Obsolete”, The Artist Economy (Substack)

Dispatch
Drawing on an interview with Jimmy Iovine, Joel Gouveia argues that music streaming is built on a structurally flawed business model: services like Spotify pay out roughly 70% of revenue to rights holders, so costs rise almost linearly with each new user instead of benefiting from classic tech‑style scale economics. At the same time, streaming platforms have allowed music to become fully commoditised; every service carries essentially the same catalogue, turning songs into “tap water”, where the only real differentiation lies in price, interface and playlists, and where device and platform owners capture most of the strategic power. Against that backdrop, Gouveia points to the long‑term value of artists owning their communities and building higher‑margin offerings around a smaller base of committed fans. A set of lessons that feel increasingly relevant for news publishers navigating an era of platforms, AI‑summaries and licensing battles, and searching for deeper, more durable relationships with readers rather than just more reach.

Reflections
1. If making content infinitely accessible at a low flat price tends to commoditise it, how should news organisations design bundles and pricing so that journalism does not become the informational equivalent of tap water?
2. What would a “direct‑to‑reader” strategy look like for news, where a smaller but committed audience generates more sustainable economics than a mass of anonymous, low‑value users funneled through platforms?

III. On the Frontlines of Trust: Press Freedom and Information War

Signal 11.8 – “Holding the line: Hungary’s independent media and the quiet work of neutrality”

Source: András Pethő, “How Hungary’s Independent Media Held the Line”, Nieman Reports

Dispatch
In Nieman Reports, András Pethő describes how a cluster of Hungarian newsrooms survived 16 years of Viktor Orbán’s media capture not by becoming an opposition in exile, but by stubbornly remaining independent watchdogs in a system that increasingly rewarded loyalty. As public broadcasters and major private outlets were brought under political and business control, journalists who refused to align either left captured newsrooms or founded new, often small and reader‑supported investigative outlets like Direkt36, focusing on methodical reporting into corruption, cronyism and abuses of power. Their work, from long‑form investigations to documentaries watched by millions, did not campaign for a particular party; instead, it preserved one core democratic function that other institutions had largely abandoned: documenting facts about how power was used, and making that record available to anyone, including citizens who later chose to vote Orbán out. The Hungarian case quietly underlines a deeper point for journalism everywhere: in hard times, the most radical stance is often not partisan resistance, but the calm insistence on neutrality, verification and independence as a craft. This is also the very DNA of news agencies as infrastructures that are meant to serve the society, regardless of who governs.

Reflections
1. When political pressure intensifies, how can news organisations distinguish clearly between being independent and being “the opposition” – and communicate that distinction to audiences who may see everything as partisan?
2. In countries where formal checks and balances are weakening, how might news agencies and other neutral infrastructures prepare to “hold the line” so that a factual record of power remains available, even if politics swings sharply in one direction or another?

Signal 11.9 – “When the state points, others pull the trigger”

Source: “Press freedom in Latin America receding in face of state violence and organized crime”, LatAm Journalism Review

Dispatch
LatAm Journalism Review, citing the Voces del Sur “Shadow Report”, describes a press‑freedom landscape where journalists in Latin America are squeezed between two converging threats: organised crime on the one hand and increasingly aggressive state actors on the other. Over one year, the report documents 3,766 attacks in 17 countries, including 14 journalists killed, with nearly half of all aggressions attributed to public officials, police or security forces who harass, prosecute or assault reporters and systematically use stigmatizing speeches that frame the media as enemies. This climate of official hostility, combined with direct violence from criminal groups, pushes newsrooms into self‑censorship and withdrawal from entire regions, creating “information deserts” where citizens lose access to reliable coverage of corruption, crime and public policy; a dynamic that not only weakens democracy as an abstract ideal, but achieves what both abusive authorities and criminal networks want: silence.

Reflections
1. When elected officials routinely stigmatise journalists in their speeches, at what point does that rhetoric become a form of state violence in itself, even before a single physical attack occurs?
2. In countries that still see themselves as safe for journalism, what concrete infrastructure guarantees are actually in place to prevent a similar pattern from emerging if political rhetoric were to harden against the press?

Signal 11.10 – “Memes, slop and war: when propaganda learns our algorithms”

Source: Institute for Strategic Dialogue (ISD), “How pro-Iran networks gained a billion views on war propaganda”

Dispatch
ISD documents how two small, loosely coordinated networks on X amassed more than a billion views in the first month of the Iran conflict by packaging pro‑Iran war narratives in the most engagement‑friendly formats the platform rewards: memes, humour and fast‑produced “AI slop” that blurs the line between news, parody and spectacle. Fewer than 50 accounts, many of them paying for verification, repeatedly amplified each other’s posts and exploited X’s recommendation system so that fabricated attacks, fake videos and exaggerated battlefield claims were surfaced to vast audiences simply because they generated clicks and reactions, not because they were credible. ISD is cautious about attributing direct state control, but the effect is clear: actors aligned with Iranian war narratives have learned to invert a system built for commercial virality and turn its core weakness, an addiction to engagement, into a force multiplier for propaganda at scale.

Reflections
1. What does it mean for democratic societies when the very features designed to maximise engagement on commercial platforms become the easiest levers for militarised propaganda to reach hundreds of millions of people?
2. If adversarial actors are now systematically optimising for our recommendation algorithms, what infrastructural safeguards, beyond individual media literacy, are needed to ensure that our public opinion is not quietly subordinated to whoever best plays the engagement game?

Signal 11.11 – “When connectivity becomes a weapon against trust”

Source: Konstantinos Komaitis, “The real target of Russia’s internet strategy isn’t infrastructure—it’s trust”, DFRLab

Dispatch
In his DFRLab essay, Konstantinos Komaitis argues that Russia’s external internet strategy is not primarily about taking down Western cables, clouds or platforms, but about something more subtle: using those same networks to erode people’s trust in everything that runs on them. Through long‑running disinformation campaigns, amplification of polarising narratives and systematic efforts to cast doubt on elections, independent media, public institutions and even the security of digital infrastructure, Russian operations try to ensure that no single version of reality feels fully credible. Rather than aiming for one dominant story, the goal is to flood the environment with enough conflicting claims, hoaxes and “just asking questions” content that citizens begin to treat all information as equally suspect, weakening the connective tissue of Western democracies and economies, which depend on shared confidence in communications networks, cloud services and critical information providers.

Reflections
• If the central objective of hostile information operations is to make everything feel untrustworthy, how should democracies rethink “resilience” beyond firewalls and uptime to include the health of public trust itself?
• What role can critical information infrastructures such as news agencies play as a stabilising layer of verification when entire information environments are being engineered to feel chaotic and relative?


IV. On AI licensing, AI Discovery and Living with AI in the News Ecosystem

Signal 11.12 – “Reuters, AI and the price of live news”

Source: Charlotte Tobitt, “Thomson Reuters boss says AI licensing deals only involve archive text”, Press Gazette

Dispatch
At the Truth Tellers Summit in London, Thomson Reuters CEO Steve Hasker explained that the company has so far struck only a small number of AI licensing deals, limited strictly to its text archive and explicitly excluding the live news file, images, video and audio. The strategy is to charge the highest possible price for archive text and to keep contracts short, in order to both fund “the next generation of journalists” and retain leverage in a market Hasker expects to be more disruptive than the arrival of the internet, Google or social media. At the same time, he describes a stark imbalance: major chatbots can still surface Reuters‑sourced breaking news without paying “a penny”, reflecting what he calls a Silicon Valley mindset of “if I can scrape it, it’s fair use”, and raising the question of how even a global wire can sustain live newsgathering if its most valuable, constantly refreshed content is treated as free fuel for AI products. If Reuters struggles to secure sustainable terms for real‑time journalism in this environment, the situation looks even more precarious for smaller agencies, which strengthens the argument that newswires should be recognised and protected as part of critical information infrastructure, not just another content supplier in licensing negotiations.

Reflections
1. If a leading global agency can only license its archives on acceptable terms, what realistic options remain for adequately valuing and protecting live news in AI‑driven information ecosystems?
2. What does it take for legislators and regulators to treat news agenciesnot as ordinary rightsholders negotiating with much larger tech firms, but as critical infrastructure whose ability to fund reporting is a matter of public interest?

Signal 11.13 – “dpa‑iq: a newswire preparing for the agentic age”

Source: Teemu Henriksson, “How the German Press Agency is reinventing news distribution for the agentic age”, WAN‑IFRA

Dispatch
Founded in 1949 as an independent, cooperative news agency owned by many German media companies, dpa is now repositioning itself for a world in which a growing share of knowledge and information work in newsrooms is done not by people directly, but by AI‑driven agents acting on their behalf. Its new platform, dpa‑iq, is explicitly designed as a trusted, API‑based information layer for this “agentic age”: instead of sending finished stories into human‑facing systems only, dpa exposes its text, photos, video, audio and data through machine‑readable endpoints so that news organisations’ own AI tools can automatically retrieve and assemble what they need. Architecturally, dpa‑iq supports multi‑source retrieval, granular rights and rate‑limiting, and integrations with tools like Langdock, OpenAI and workflow platforms, enabling use cases such as automatically generated morning newsletters or background briefings that are built on verified agency content rather than whatever the open web happens to offer. Seen from the wider perspective of critical information infrastructure, the move also points beyond media: if news agencies are to remain neutral backbones for reliable information in an AI‑mediated world, similar machine‑to‑machine access will be essential not only for publishers, but also for governments, companies and other institutions that depend on trustworthy, continuously updated news signals.

Reflections
1. If more editorial work is mediated by AI agents, how can news agencies ensure that their content remains the default “source of truth” those systems turn to, rather than being sidelined by cheaper, less reliable open‑web data?
2. What governance and transparency standards should apply when agencies open machine‑to‑machine access to governments and corporations, so that critical information infrastructure strengthens democratic accountability?

Signal 11.14 – “When AI doesn’t think for journalists”

Source: Chad Davis, “AI, discovery and media’s next phase”, Editor & Publisher

Dispatch
Chad Davis argues that generative AI should be seen as “the forge, not the fire”: a powerful way to prototype and iterate, but not a substitute for editorial intent or human‑to‑human thinking. AI can rapidly reach an average outcome and make generation extremely cheap, yet in journalism success is still measured in trust, not in speed or technical “magic”. In a world where token costs have collapsed, the real differentiation for news organisations lives in their workflows and standards — in transparent processes, clear motives and high bars for ethics, reporting and fact‑checking. Davis suggests that, as AI becomes a dominant discovery layer, publishers will increasingly have to make a conscious choice with each piece: am I optimising this primarily for human reading, for machine reading, or deliberately for both?

Reflections
1. What does it look like to build parallel publishing paths, optimised for human narrative and for machine discovery?
2. How can news agencies and publishers communicate to audiences that AI helps with production, but that the core work of judgement, accountability and trust remains firmly in human hands?

Signal 11.15 – “From scraping to standards: Mediahuis backs SPUR”

Source: Journalism.co.uk, “Mediahuis joins SPUR Coalition to help set global AI standards for journalism”

Dispatch
Journalism.co.uk reports that Mediahuis, one of Europe’s major news groups and a shareholder of Belga News Agency, has joined the SPUR Coalition (Standards for Publisher Usage Rights) alongside founding members such as the BBC, Financial Times and Guardian Media Group. SPUR aims to move the industry beyond ad‑hoc scraping and opaque “fair use” claims by developing common technical tools and principles for how AI companies may access, use and remunerate journalism, from training and citation to live answer products. Mediahuis CEO Gert Ysebaert calls this “a defining moment” in which publishers must work together to ensure that quality journalism is used “responsibly” in AI systems and that there is a “fair and transparent value exchange” between tech platforms and newsrooms. For news agencies that still depend heavily on revenues from exactly these publishers, the move is more than symbolism: it proves that the industry is motivated to work out AI‑era business rules that strengthen journalism and it's basic structures like news agencies.

Reflections
1. Under what conditions could SPUR realistically function as a “NATO for news” in AI negotiations, and what early wins should publishers look for to test whether coalitions like SPUR can deliver tangible benefits, rather than becoming another well‑intentioned industry talking shop?
2. How can members like Mediahuis, the BBC, the Financial Times and the Guardian translate shared principles into concrete, enforceable standards that tech companies actually adopt in their products and APIs?


Signal 11.16 – “Fighting AI is a losing battle”

Source: “Demi Moore calls fighting AI a losing battle ahead of Cannes opening”, Reuters

Dispatch
On the eve of the Cannes Film Festival opening, actor and jury member Demi Moore told Reuters that “AI is here” and that “to fight it is… to fight something that is a battle that we will lose”, urging the film industry to find ways to “work with” artificial intelligence rather than simply resist it. With both feet on the ground, she framed AI less as an abstract threat and more as a practical reality check, immediately adding that her instinct is to say the industry is “probably not” doing enough yet to protect creators from misuse of their image, voice and work. Coming from the jury steps at Cannes rather than a tech panel, her intervention lands as a broader question for culture and media: are sectors that depend on human creativity and public trust actually prepared for a world in which AI is a permanent co‑presence?

Reflections
1. If “fighting AI” in the sense of denying its existence is indeed a losing battle, what does a realistic, protective stance look like for film, news and other cultural industries that rely on human performance and authorship?
2. How can institutions like film festivals, news organisations and news agencies use their symbolic and market power to set norms around responsible AI use?