Dispatch 30
Disinformation works as an industry. It has suppliers, subcontractors, distribution budgets and a return on investment, and it sells its product far below the cost of the truth. Tomorrow in Brussels, Alliance4Europe brings researchers, fact-checkers, journalists and policymakers together for its annual Counter Disinformation Network conference, under the title Behind the Influence. The programme targets the infrastructure behind individual falsehoods: the networks, accounts, funding and platform weaknesses that allow one lie to reach millions of people at almost no cost.
Today is also World News Day. Verification, field reporting and correction require trained people, money and time, and AFP’s global news director reminded his readers of exactly that this week. The economics of disinformation therefore meet the economics of journalism. One side produces falsehood cheaply and at scale; the other produces verified facts slowly and at a real cost.
This week’s links approach that tension from several angles. European institutions and civil society build tools to disrupt influence operations. Press agencies in Asia and Europe propose shared verification services. Legislators from Brasília to Manila try to protect elections without handing governments the power to decide what counts as true. Meanwhile, AI answer engines change how audiences find news, which raises two questions: who answers for the content of an AI answer, and who pays for the facts inside it.
Eleven signals in five chapters, from influence operations to licensing and liability.
Europe as an information network under pressure. The red nodes and connections represent coordinated influence operations across borders and platforms; the blue connections represent trusted information, verification and cooperation. The aim is to understand the networks behind individual falsehoods and to disrupt the infrastructure that allows disinformation to scale. (Synthetic Image)
I. On disrupting influence operations
Brussels hosts the European counter-disinformation community on 28 and 29 September.
30.1 — Alliance4Europe brings a disruption toolkit to Brussels
Alliance4Europe hosts the Counter Disinformation Network 2026 Conference in Brussels on 28 September, followed on 29 September by a side event in the European Parliament on coordinated networks from West Africa and Southeast Asia that exploit X during European elections. The programme covers foreign information manipulation and interference (FIMI), election integrity, influence campaigns against strategic sectors such as renewable energy, and the European launch of the DISRUPT Toolkit. The organisation’s Counter Disinformation Network connects more than 110 member organisations and 450 practitioners, and it sends incident alerts to more than 300 recipients.
The toolkit helps practitioners act on what they find. Its open DISRUPT Framework offers shared definitions and measures, from the preparation of evidence to disruption and mitigation, and its workflows cover six types of influence operations in four EU countries. Alliance4Europe built it after consultations with more than 170 practitioners. For newsrooms, one session deserves particular attention: a closed roundtable led by the Fojo Media Institute on how to report on influence operations accurately without amplifying them, a dilemma every editor knows who has ever published a story about a fake story.
The approach treats disinformation as a system with costs, suppliers and weak points. A common vocabulary allows a researcher in Warsaw, a fact-checker in Bucharest and a platform trust team to describe the same operation in the same terms, and that shared language makes coordinated action possible. Europe already has many assets to defend its information space; the toolkit tries to make them work together.
Reflections
What role can a newsroom accept in a whole-of-society response before its independence comes into question?
How can editors report on an influence operation so that the public learns what happened and the operators gain nothing from the coverage?
30.2 — Storm-1516 and hired networks target Europe’s elections
A Bloomberg investigation counts more than 190 fabricated stories from the Russian operation Storm-1516 since August 2023, built on forged documents, staged testimonies and AI-manipulated video. More than 40 percent of those stories aim to weaken support for Ukraine and a third target elections; Western officials attribute the operation to Russia’s military intelligence service, the GRU. In April, Euronews showed the same group at work in Hungary, where a fabricated claim accused opposition leader Péter Magyar of plans to reintroduce military conscription.
A second model relies on rented reach. Alliance4Europe, Trollrensics and SVT followed one network across the Dutch elections of 2025, the Hungarian elections of 2026 and the Swedish general election this September, with 8,505 suspected accounts in Benin and further activity linked to other countries in West Africa and Southeast Asia. The researchers report that X restricted or removed fewer than one percent of the accounts they had logged earlier, and they raise the platform’s obligations under Articles 34 and 35 of the Digital Services Act. Public broadcasters meet the same technique on their own pages: Eurovision News Spotlight analysed 17.5 million comments under posts from ORF, RTVE, RFI, Tagesschau, ZDFheute and others, and found coordinated networks that manufacture debate in the comment sections of trusted media.
The Guardian describes how state actors and domestic influencers combine deepfakes, fake documentaries and AI-generated content to deepen division and despair. For every news organisation, the pattern carries a practical lesson: influence operations reuse their infrastructure across countries, rent it out to clients, and target the audience spaces that quality media built over years. The comment section under a public broadcaster’s post has become contested territory, and the broadcaster carries responsibility for it.
Reflections
When one hired network can serve three elections in two years, which actor should bear the cost of detecting it?
How far can a public broadcaster moderate its own comment sections before moderation turns into a question of editorial freedom?
30.3 — The European Parliament asks for a Centre for Democratic Resilience
The European Parliament adopted the report of its Special Committee on the European Democracy Shield by 420 votes to 220, with 26 abstentions. MEPs name Russia as the primary external threat, alongside Belarus, China, Iran and North Korea, and they support an independent EU Centre for Democratic Resilience with its own mandate, budget and parliamentary oversight. The report also calls for stronger protection of media freedom, action against bot-driven electoral interference and rigorous enforcement of the Digital Services Act and the AI Act.
The EPRS briefing sets out the background. The Commission and the High Representative presented the European Democracy Shield in a joint communication of 12 November 2025, and the special committee now recommends a binding legal act and a dedicated budget for the Centre. In legal terms, Parliament adopted an own-initiative report, and the Commission holds the right of initiative for any legislation. Press agencies and publishers should follow the design of the Centre closely, because its governance, its funding and the role it gives to independent media will determine whether it strengthens existing verification work or duplicates it.
Rapporteur Tomas Tobé summarised the political line in one phrase: “foreign interference is not free speech”. Europe now combines a regulatory layer in the DSA and the AI Act, an institutional layer in the proposed Centre, and a civil society layer in networks such as Alliance4Europe. The task for the coming year concerns the coordination among those three layers, and the place of independent journalism inside that architecture.
Reflections
Which safeguards should a European Centre for Democratic Resilience build in so that editors never need its approval for their work?
How can public funding for counter-disinformation reach independent newsrooms without creating a new dependence?
II. On press agencies as verification infrastructure
Agencies propose to organise verification as a shared, cross-border service.
30.4 — Asia-Pacific agencies propose a joint fact-checking task force
ANTARA’s president director Benny Siga Butarbutar called on the members of the Organisation of Asia-Pacific News Agencies to build cross-border mechanisms for fact-checking and verification. In a second ANTARA report, he proposes a permanent task force for the 41 OANA members from 35 countries, with rapid verification networks, early warning, shared forensic metadata, multilingual media literacy, common AI ethics standards and fair licensing cooperation with AI companies. ANTARA is Indonesia’s state news agency, and these proposals express its own institutional position.
In Europe, AFP leads CLIC, a project that the European Commission co-finances, together with France Médias Monde and the Slovenian newsroom Oštro; the partners work with well-known digital creators on videos, carousels and podcasts that counter disinformation during election cycles in France and Slovenia. In a column for Dawn, as UA.NEWS reports, AFP’s global news director Phil Chetwynd describes what verified reporting costs: AFP sent eleven journalists to the floods on the Nepal-Tibet border, one photographer needed thirteen hours to reach the main disaster zone, and more than a hundred AFP journalists work full-time on verification and fact-checking. At the ASEF Journalists’ Seminar in Bangkok, 22 journalists from Asia and Europe discussed the same combination of AI, verification and resilience.
Agencies hold a specific position in the information system. They produce verified facts at wholesale level for publishers, broadcasters, public authorities and, increasingly, for machines. When agencies pool verification across borders, they create a common good with a real cost base, and someone has to finance that cost base. There is no such thing as a free fact.
Reflections
Which parts of verification can agencies share across borders, and which parts belong to national editorial judgement?
How can a joint task force of agencies with very different ownership models earn and keep the trust of the public?
30.5 — Newsrooms treat verification as forensic work
Columbia Journalism Review examines the future of visual investigations now that AI can generate and manipulate images, access to satellite and user-generated material shrinks, and newsrooms reach different conclusions from the same footage, as they did after the explosion at Al-Ahli Arab Hospital in Gaza in October 2023. Experts from Bellingcat, Forensic Architecture, the New York Times and the Financial Times argue for combining visual analysis with reporting, witness interviews and data. Journalists also borrow tools from the legal profession: in a press release, Everlaw reports that newsroom investigations on its platform tripled in one year, with more than 50 newsrooms and organisations including USA TODAY, Le Monde and WIRED.
The same technology also produces errors inside newsrooms. ABC Media Watch documents how local Herald Sun Leader titles in Melbourne published council stories with AI-generated errors, including a quote from a resident at a meeting she did not attend; News Corp describes the project as an experiment of 12 to 18 months in communities with reduced coverage, with human review. For journalists in exile, technology offers reach and security and also creates dependence on large platforms, as speakers at ISOJ told LatAm Journalism Review; RSF counts about 1,500 journalists from more than 60 countries in exile between 2021 and 2024.
Verification has become a discipline with its own methods, tools and standards of proof, closer to forensic work than to traditional desk editing. The quality newsroom carries that discipline, and it applies it to its own output as strictly as to external claims. A newsroom that automates reporting without verification produces the same kind of material that fact-checkers spend their days correcting.
Reflections
What standard of proof should a newsroom require before it publishes a visual claim from a conflict zone?
When a publisher tests AI in communities with reduced coverage, who in the organisation answers for the errors?
III. On elections, deepfakes and the limits of law
Legislators on four continents look for a balance between protecting elections and protecting speech.
30.6 — Election campaigns test labelling and disclosure rules
Ahead of Brazil’s first round on 4 October, RSF considers the measures of the Superior Electoral Court largely inadequate, despite labelling requirements and a ban on candidate deepfakes. A fake video of Jornal Nacional anchor Renata Vasconcellos collected more than 420,000 views on TikTok before the platform removed it. In the United States, AFP reports that the Wesleyan Media Project identified at least 164 political ads this year that campaigns generated or enhanced with AI, at a cost of just under 80 million dollars, and Governor Gavin Newsom signed a California law that allows regulators to fine influencers up to 5,000 dollars per violation when they fail to disclose paid political content.
India’s public broadcaster Prasar Bharati approaches the question through procurement. Its draft call for AI suppliers requires statutory labels, machine-readable provenance such as C2PA, on-screen AI disclosure, consent for the cloning of voices or likenesses, and human editorial review. That approach deserves attention from every media organisation that buys synthetic content. A contract can impose provenance and disclosure faster than legislation, and it binds the supplier directly.
The Brazilian case also shows how fakes borrow the credibility of trusted anchors and news brands. Every deepfake of a presenter exploits a reputation that a newsroom built over decades, which turns provenance into a matter of brand protection for publishers and broadcasters. Labelling rules help; provenance data that accompanies the content itself offers the more durable safeguard.
Reflections
If a deepfake uses the face of a news anchor, which obligations fall on the broadcaster, the platform and the electoral authority?
How many media buyers would need to require C2PA provenance before it becomes the market standard?
30.7 — Africa and Asia weigh information integrity against free expression
In Addis Ababa, IGAD’s early-warning mechanism CEWARN brought together the African Union Commission, the office of the UN Special Envoy for the Horn of Africa and UNDP to establish a national multi-agency platform against hate speech and disinformation, with pilots during electoral cycles in Ethiopia and South Sudan. The Media Foundation for West Africa asks the African Union for a binding continental framework that addresses disinformation while it protects independent media and access to information. In South Africa, Moxii Africa, Sanef and Africa Check prepare the local elections with a partnership that combines fact-checking, journalist safety and a draft code for election journalism.
The Philippines shows the risk on the other side. Human Rights Watch warns that the proposed Digital Media Anti-False Information Act, which passed second reading in the House in May, contains vague provisions, sets prison terms of six to twelve years for publishing false information, and gives the ICT department broad powers over platforms. A law that lets a government define falsehood also gives that government a tool against its critics.
These initiatives share one dilemma: every mechanism against disinformation needs someone who decides what counts as false. The strongest models spread that decision across independent fact-checkers, editors’ forums and courts, and keep it away from a single ministry. Europe’s approach through platform obligations and independent networks offers one reference; the African proposals for a rights-based continental framework offer another.
Reflections
Who should hold the authority to label information as false, and who should review that decision?
What can European regulators learn from African proposals that put access to information and media independence at the centre?
IV. On audiences in the age of answer engines
Audiences increasingly find news through AI interfaces, and publishers invest in direct relationships.
30.8 — Search referrals fall as AI answers grow
Chartbeat data, as Search Engine Journal reports, show a 40.2 percent year-on-year decline in Google Search traffic to publishers between July 2025 and July 2026, after a decline of 21.9 percent the year before; traffic from Google Discover fell by 34.3 percent. SparkToro’s clickstream data indicates that 68 percent of Google searches in the United States end without a click to the open web. Comscore sees AI use spread across providers, with ChatGPT’s share of prompts down from 70 to 50 percent in six months, Gemini at 30 percent and Claude at 11 percent. In the United Kingdom, Ofcom finds that 21 percent of adults used an AI service for news in the past month, while 56 percent trust news that AI writes less than news that a person writes.
Machines also generate a growing share of the traffic that remains. News UK’s chief technology officer told Press Gazette that non-human visitors accounted for about a fifth of the publisher’s traffic over six months, a third of them malicious, and that News UK switched from a block list to an allow list for AI crawlers. The Wikimedia Foundation reports to Adweek an 8 percent decline in human visits while bot traffic strains its servers, and it now develops an enterprise API for AI companies. The Comscore data adds a telling detail: Tripadvisor supplied 61 percent of the travel content that AI platforms used, and received 21 percent of the citations.
The economic consequence follows. When a machine replaces the visitor, value creation depends on the point of access, and whoever controls access for the machine sets the terms of the new value chain. Allow lists, enterprise APIs and licences express the same principle in three forms. Publishers that register, authenticate and price machine access keep a position in that chain.
Reflections
What share of a publisher’s audience strategy should still depend on referrals from search?
How should a news organisation measure its reach when AI systems use its content far more often than they cite it?
30.9 — Publishers invest in direct relationships and verifiable standards
Guardian chief executive Anna Bateson told The Rebooting that the Guardian answers AI with direct human relationships, podcasts, newsletters, video and engagement time as a central metric; the organisation generates 282 million pounds and expects licensing, including agentic licensing, to gain importance. FT chief executive Jon Slade told Press Gazette that digital sources now provide 80 percent of the FT’s 540 million pounds in revenue, and that its AI deals with OpenAI and Google include “red button” clauses that allow the FT to stop live feeds. Schibsted’s Svenska Dagbladet grew from 98,000 to more than 158,000 digital subscribers in two years, and The Atlantic’s chief executive Nicholas Thompson reports 1.6 million subscribers and an OpenAI deal with contractual limits on summarising, quoting and imitating its authors.
Inside newsrooms, the same investment takes the form of skills and standards. As WAN-IFRA describes, the Hungarian news site 24.hu turned one editor’s tool for tracking politicians’ social media into a newsroom-wide AI programme in which a journalist checks and approves all AI-assisted material before publication. Editor & Publisher obtained the first Ethical AI Certification from the Alliance for Audited Media, which assesses AI policies, disclosure, human oversight and risk management. Mediahuis relaunches Hyves, the Dutch social network, with a feed of friends’ posts only, no addictive algorithms and data on European servers, a design choice that also leaves less room for the amplification tactics of the first chapter.
As the New York Times marks its 175th anniversary, A.G. Sulzberger recalls some 15 million articles and the loss of nearly 3,500 American newspapers in two decades. The organisations that perform well in this environment share a few traits: a strong brand, a direct relationship with readers, contractual control over machine use, and standards that an outsider can audit. In each case, the quality newsroom and its collective reputation provide the underlying asset.
Reflections
Which of these assets can a mid-sized European publisher build on its own, and which require cooperation with others?
What would an external audit of a newsroom’s AI practices need to cover before readers accept it as credible?
V. On licensing and law in search of equilibrium
Lawsuits and licences together set the boundaries and the price of news for AI.
30.10 — AI licensing starts to price news by use
Adam Greenberg of the New York Times told Digiday that AI content marketplaces remain “underdeveloped”, and that the Times assesses deals on value exchange, the sustainability of the partnership, and control and guardrails; its partnerships team has doubled to ten people. Press Gazette’s tracker records OpenAI’s first deals in India, with Bennett, Coleman and Co, publisher of the Times of India, at a reported 5 million dollars a year and the Indian Express Group at 3 million dollars, while the Digital News Publishers Association sues OpenAI for copyright infringement before the Delhi High Court.
The payment models change as well. According to Search Engine Journal, Google now tests payments for content that contributes significantly to answers in the Gemini app, AI Overviews and AI Mode, and it has approached dozens of publishers. K-EnterTech Hub describes a wider trend at Google and Meta towards usage-based fees and revenue sharing. In Germany and Austria, NewsGuard AI answers news questions only from some 12,000 sources that NewsGuard has rated for credibility and transparency, and it shares revenue with the media it cites.
The market now tests two logics: payment for access to a collection, and payment per use of a specific piece of content. News production carries a fixed cost base, and publishers cannot audit what happens with their content after delivery; both facts favour access as the billable moment, with usage data as a basis for adjustment. NewsGuard AI adds a third element, because it makes credibility a selection criterion and pays for it. Trust itself becomes a priced input.
Reflections
At which moment should a news organisation bill a machine: at access, at use, or at citation?
If credibility becomes a selection criterion for AI answers, who should certify it?
30.11 — Courts and governments set boundaries for AI answers
The Regional Court of Munich I, in preliminary proceedings, ordered Google to stop showing an AI Overview that linked a Munich publishing house to scams and subscription traps. The court treated the AI Overview as Google’s own content, outside the safe harbour that protects search engines for statements of third parties, and Google announced an appeal. In Lawfare, Ignacio Cofone argues that this reasoning, together with the case law of the Court of Justice and the revised Product Liability Directive that applies from December 2026, could expose AI providers to liability for false outputs.
In Australia, Reuters reports that Anthropic and OpenAI ask the government to allow limited and conditional training on Australian creative content, after Canberra ruled out a broad copyright exception; a parliamentary committee reports in November. The European Union answered part of that question with the text and data mining provisions of the CDSM Directive, which allow rightsholders to reserve their rights. At the international level, 22 countries back a declaration that calls for transparent safety protocols, pre-deployment testing and independent evaluation of AI, with Finnish President Alexander Stubb as its main promoter; the United States and China have not signed.
Liability and licensing belong to the same process. Courts establish who answers for an AI output and which uses require permission; negotiations then set a workable price within those boundaries. The Munich ruling adds an important principle from a European legal perspective: a system that formulates the answer also carries responsibility for it. That principle connects this chapter with the first, because it applies to every actor that distributes claims at scale.
Reflections
If AI providers carry liability for their answers, how will that change their appetite for licensed and verified sources?
Which boundaries should courts draw first, and which can negotiations settle?
Conclusion
The title of tomorrow’s conference, Behind the Influence, also names the structural question of this week. Every piece of disinformation relies on an infrastructure: accounts, networks, budgets, hosting and platform rules that someone designed and someone pays for. Every verified fact relies on another infrastructure: correspondents, fact-checkers, editors, archives and agencies. Both carry costs. The first shifts its costs onto society; the second needs readers, clients and licensees to cover them.
AI answer engines connect the two. They draw on both kinds of material, and they decide which one reaches the audience. The Munich court, NewsGuard’s source selection and the licensing terms of the New York Times, the FT and The Atlantic all address the same point: whoever formulates an answer takes responsibility for its accuracy, and accuracy has a price.
For Europe, the task consists of connecting the assets it already has. The Democracy Shield offers the institutional frame, the DSA sets the obligations for platforms, civil society networks provide detection capacity, and press agencies and quality newsrooms supply verified content. Each part works better when the others function. This week, Asia-Pacific agencies proposed a comparable model for their region, and African media organisations asked for one that starts from freedom of expression.
A modern economy relies on a cadastre, the public register that records who owns a property, who may use it and on what terms. Without that register, every transaction turns into a dispute. Information needs the same kind of record: who produced a fact, who verified it, and who may use it under which licence. Provenance standards, agency verification and licensing contracts together form that register. Disinformation profits wherever the register is missing, and journalism keeps its value wherever the register holds.
On World News Day, that defines a fitting task for the sector: keep the register, and control how the machines read it.
Patrick Lacroix writes Cyber Territories in a personal capacity. The author uses AI tools for research and drafting and retains full editorial responsibility.