OpenAI Bans Russian and Iranian ChatGPT Accounts Behind Fake Journalists and Planted Stories

OpenAI said Thursday that it banned clusters of ChatGPT accounts traced to Russia and Iran for supporting covert influence operations. One of them, a Russian campaign, drew public comments from politicians, which led OpenAI to rate it category 5 out of 6 on the Breakout Scale, according to The Decoder. That is the first category 5 case in two and a half years of reporting.

OpenAI says both campaigns were used to "launder geopolitical, conflict-related messaging into their target audiences," with ChatGPT producing both internal and public-facing content, according to CNA. The company describes "false front" operations that pair AI with conventional deceptive techniques. The model did not do the lying alone. It had human help and a good costume department.

This is not OpenAI's first such report. In June 2026 it said it disrupted two Chinese-linked operations, one focused on data centers and one aimed at criticizing U.S. tariffs.

Russia: a research center with real employees

The Russian operation is called "Dark Clark," after a fictitious persona the operators used. According to The Decoder, it spread disinformation across Latin America to discredit Ukraine and destabilize local politics.

The more interesting part is the staffing. The operators ran a front organization called the Social Research Center (SRC), which brought in local staffers in Latin America to do research and produce content. OpenAI says "the available evidence suggests that these on-the-ground employees had no idea that they were working for a Russian operation, and conducted research tasks in good faith." So the fake think tank had real employees, and the employees were the part that was not fake.

Fabricated audio files and documents triggered fact-checks and official denials in Ecuador and Peru, according to The Decoder. The report also describes alleged attempts to spread fake leaked audio in Bolivia.

One detail deserves a footnote. Internal reports showed the operators asked OpenAI models to help identify incidents in target countries that they could take credit for, "even if they had not been involved." That is a request for plausible-sounding credit, and it tells you what the operation was optimizing for: the look of influence, not necessarily the substance.

Iran: seven bylines, nearly 100 articles

The Iranian operation is called "Bogus Bylines." It created seven fake journalist personas posing as Western or American reporters covering foreign affairs and geopolitics. Between July 2025 and October 2026, those personas placed or syndicated nearly 100 articles across roughly a dozen small-to-medium online outlets, according to the AFP account of the report. The articles concerned the US-Iran conflict.

The outlets that published the fake authors included the progressive U.S. news site Daily Kos and the U.K.-based Middle East Monitor. OpenAI's full list of outlets is not in the coverage reviewed here.

The personas also left some trail. One, "Ervin B. Hoskins," had profiles on X and Instagram. According to screenshots provided by OpenAI, X's transparency information showed the account connected via a "West Asia android app," and Instagram's showed it was based in Iran. A Western freelancer who files from a West Asia android app is, let us say, a stretch.

OpenAI rated the media pitching and publication strand of the operation Category 4. It noted that articles were at times promoted by the outlets' own social accounts, "giving the Iranian operators a potential way to access audiences that they would not otherwise have reached." The social media comments in English and Persian about the U.S.-Iran war did far worse: typically single or double-digit likes, views and replies, with many coming from the operation's own accounts. OpenAI says the activity looked consistent with a hired commercial actor.

The numbers, and the gaps

Operation	Origin	OpenAI rating	What it did
Dark Clark	Russia	Category 5 of 6	Fake front think tank, fabricated audio and documents in Latin America
Bogus Bylines	Iran	Category 4 (media strand)	Seven fake journalists, nearly 100 articles

Now the part I read first: what was not disclosed. OpenAI did not disclose how many accounts it banned, and it did not identify the ultimate operators behind either campaign. The coverage also does not name which politicians reacted to the Russian content or what they said. Reports differ on where they were: NPR reported that internal reports showed a response from a member of the European Parliament, while The Decoder, CNA and AFP emphasize fact-checks and denials in Latin America, such as Ecuador and Peru.

The category 5 rating also needs reading with care. By OpenAI's own description it reflects politicians reacting to the content. That measures reach into public debate, not whether anyone was persuaded. Likewise, the Iranian strand's Category 4 comes with a note that outlets promoted the articles, which says something about the outlets' vetting and not just about the model.

What to watch

Three things. First, whether OpenAI puts a number on the banned accounts. "Clusters" is not a head count. Second, whether the outlets that ran the fake bylines say anything about how the articles got through. Third, whether the next report finds another category 5 or this one remains a one-off. The baseline cited here is OpenAI's own reporting history: two and a half years.

The operators used ChatGPT to create internal and public-facing content. The Iranian campaign also pitched articles to news outlets. The weakest link, as ever, was the part with a byline and an editor.