AI fake journalists backed by an influence operation managed to place articles in genuine online publications, according to an investigation released by OpenAI on 8 October 2026. The company says it banned two covert campaigns, one originating in Iran and another in Russia, that used artificial intelligence alongside established deception techniques. The worrying detail is not simply that the text may have been AI-assisted. It is that readers encountered articles in real outlets without knowing who was behind the supposed writer.
The OpenAI investigation describes Iranian operators using seven journalist identities to pitch long articles, while a Russia-origin campaign allegedly controlled a research organisation staffed partly by people unaware of the real operation. OpenAI reports finding almost 100 articles published or syndicated under bylines associated with the Iranian campaign. Those findings come from the company that investigated activity on its services. It distinguishes several independently observed outcomes from other claims made by the operators themselves.
How the AI fake journalists approach worked
A fabricated journalist identity can do more than attach a false name to a piece of text. It can come with a biography, social-media accounts, an apparently plausible area of expertise and a polished email to a commissioning editor. The Iranian operation asked AI tools to review articles against publications’ submission requirements and draft pitches. A busy editor might then receive what appeared to be a normal freelance approach from someone who wrote fluently and presented a credible portfolio.
This is a different problem from an obviously false social-media post. The publication receiving the material is a real organisation, and its ordinary distribution channels can lend unfamiliar writing a degree of legitimacy. The failure, when it occurs, is in verification of the contributor’s identity and interests. AI can make the presentation look professional, but it does not provide a genuine reporting history or a transparent account of who commissioned the work.
OpenAI says some false journalist personas maintained accounts on several social platforms. Its investigators compared prompts, publishing activity and observable online traces. That does not establish that every article published by the affected outlets was generated entirely by AI. In the company’s account, the technology was frequently used for revising, translating, polishing and pitching material as well as producing text.
Why real publication matters more than fake comments
The Iranian operation also generated social-media comments, often responding to coverage of the conflict involving Iran, the United States and Israel. OpenAI says the comments it could identify generally gained little engagement. The publishing effort was more consequential because material appeared in third-party publications under deceptive bylines. The company’s assessment therefore differs by distribution method, rather than assuming every generated post reached a large audience.
The Russia-origin campaign used another route. OpenAI describes what appeared to be control of a research centre in Latin America, with local workers apparently unaware of their alleged Russian backers. The operators also claimed to circulate forged documents, invented audio material and misleading stories. Independent searches found examples that resembled some reported activity and drew public denials or fact-checking. Other alleged successes could not be corroborated, and the report explicitly warns against accepting the operators’ own boasts about impact.
For a reader, the distinction is essential. A detailed case study can contain verified observations, plausible attribution and claims from unreliable adversaries in the same document. The strongest story is not that AI automatically convinced everyone. It is that sophisticated false fronts found ways to enter legitimate information channels while using inexpensive digital tools to assist the process.
A believable name is not evidence of a real reporter
Some warning signs are familiar from older impersonation schemes. A writer might have an impressive biography that cannot be verified outside their own profiles, a series of accounts with thin interaction histories, or a professional identity that appears only around one set of political topics. None of these details proves wrongdoing on its own. Real freelancers also work internationally, use pseudonyms and have limited public footprints. The relevant check is the contributor’s identity, editorial relationship, evidence and disclosure of interests, not simply whether AI tools were used in drafting.
The company links its new report to a wider programme of investigations into malicious uses of AI. The archive includes previous accounts of covert influence efforts and scams, demonstrating that this is an ongoing security problem rather than a new ability that appeared overnight. OpenAI also notes that older false-front campaigns used human writers and convincing identities before modern chatbots existed.
Ordinary readers do not usually have access to an outlet’s commissioning records. They can nevertheless pay attention to sourcing, direct documentation and the difference between a quoted expert and an unattributed assertion. LiveAIWire’s earlier reporting on AI-assisted phishing and personalisation shows a related pattern: well-written, context-aware communication can be part of a deceptive operation without being a guarantee of truth.
The lesson for editors and the public
Publishing organisations should treat contributor verification as separate from text quality. Confirming contact details through independent channels, understanding conflicts of interest and checking primary documents can prevent a persuasive-looking submission from receiving authority it has not earned. Editors may also need to revisit how they vet repeated contributors, syndication arrangements and biographies, especially where outside content is published under staff-like presentation.
Readers face a different challenge because online circulation can detach an article from the circumstances in which it was commissioned. Once an article has an established publication logo and a shareable address, it is easy to treat the location as proof of provenance. A publication is evidence that the item was published there, not that every identity and assertion behind it is authentic. LiveAIWire has looked at how Britons encounter AI-related news, making the integrity of the wider news environment a direct audience concern.
The new report should not be read as a claim that the media is uniformly compromised, nor as evidence that AI-generated writing is inherently deceptive. The deceptive act is concealment of the operator and purpose. OpenAI says it has banned the accounts it identified and shared findings with relevant parties. The harder, continuing question is how publishers and readers can recognise a credible-looking voice that belongs to an organised campaign rather than the person named beneath the headline.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.