OpenAI bans two false-front operations: a fake think tank and fake bylines
Victoria Shinder2 min readOpenAI published a threat report on 8 October describing two covert influence operations it banned — one from Russia, one from Iran — that used its models alongside older techniques to run "false front" entities: a think tank and a stable of invented journalists.
Russia: "Dark Clark". A cluster of accounts originating in Russia (connecting through VPNs, since OpenAI does not serve Russia) used ChatGPT mostly to write internal reports to an unknown superior. The reports describe a "research platform" in Latin America, the Social Research Center, run through a fake persona called "Mia Clark" — hence the nickname. According to OpenAI, the operators' reports discussed pay scales, hiring and firing, and the available evidence indicates the centre's local employees did not know they were working for a Russian group. OpenAI rates it Category 5 on the IO Breakout Scale (1 to 6) — the first Category 5 operation it has disrupted since it began reporting.
The reports claim credit for specific fakes. OpenAI says it found matching press coverage, fact checks or official denials for several: a fake email from Lima's education directorate telling schools to hold Ukraine events, which then fed coverage in Peru and Poland; a fake school ceremony instruction in Ecuador that drew a rebuttal from the education minister; and a fake audio clip of a water-company worker in Bolivia during protests, denied by the utility. In other reports the operators tried to take credit for events they had nothing to do with — in OpenAI's words, an influence operation aimed at their own employers.
Iran: "Bogus Bylines". Accounts prompting in Persian used ChatGPT to check English-language articles against a specific outlet's submission criteria, then to draft the pitch email to its editor. The pitches went out under seven fake bylines, some backstopped with accounts on X, Facebook, Instagram, Threads and Medium. OpenAI identified almost 100 articles under these names in more than a dozen small and medium international-affairs outlets, from July 2025 to October 2026, rated Category 4. Batches of social media comments, some praising the operation's own articles, got no meaningful engagement.
Why it matters for anyone who publishes. OpenAI's point is that neither operation was new in kind — it compares them to the "Alice Donovan" persona and the "PeaceData" site from 2016–2020 — but AI made the editorial work cheaper. The weak point it exploited is the editor who accepts a well-written pitch from a stranger. The defence OpenAI names is disclosure: both earlier fronts stopped once exposed. For a desk taking outside submissions, the practical check is old-fashioned — verify that a contributor exists somewhere other than in their own bio.
Source: OpenAI, "Disrupting AI-enabled 'false front' operations", 8 October 2026 — https://openai.com/index/disrupting-ai-enabled-false-front-operations