Dispatch #21 - Cyber Territories
Dispatch 21
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.