OpenAI closed ChatGPT accounts run by Russian and Iranian propaganda operations

OpenAI banned the ChatGPT accounts behind two covert influence operations, one tied to Russia and aimed at Latin America, the other originating in Iran and built around fake journalists. The company disclosed the action in a safety report released Thursday, Oct. 8, 2026. In both cases, OpenAI said, ChatGPT eased the paperwork and drafting while the operations themselves worked the way pre-AI influence campaigns did.

The Russian network posed as a research institute and the Iranian one as freelance writers, and both placed material in real outlets rather than only on social media.

The Russian network and its Social Research Center

According to The Record’s account of the OpenAI report, Russian operatives used ChatGPT to help run a covert campaign across Latin America built around a fake organization called the Social Research Center. The center presented itself as a research platform on the Indian diaspora in Latin America and on India’s relations with the region, and OpenAI said Russian operatives controlled it through a fabricated identity.

The staff were real. The campaign recruited people in Latin America to do research, and OpenAI said they “had no idea that they were working for a Russian operation.” Much of what the center published was original work by those employees rather than machine-written text. The main goal was to damage Ukraine’s reputation in the region, with some activity aimed at domestic politics in Argentina and Bolivia.

The operators turned to ChatGPT for fabrication. They generated fake leaked documents and audio scripts, and some of the false claims spread widely enough to draw fact-checks and official denials. The Insight International’s summary describes a fake audio clip imitating Ukraine’s consul in Ecuador insulting Ecuadorians, and a clip released during May protests in Bolivia that falsely claimed the state water company would cut water to La Paz, which the Bolivian government denied. That summary also reports the operation as Category 5 on the Influence Operations Breakout Scale, which runs from 1 to 6, the first operation at that level OpenAI has disrupted.

Arise reported that the fabricated identity at the head of the center was named Mia Clark, and that ChatGPT also helped prepare internal reports showing decisions on staffing, salaries and recruitment.

OpenAI did not pin the campaign on a single actor. The Record reports the company said evidence suggested links to Politology, also called La Compania, which it described as a successor to influence operations associated with Yevgeny Prigozhin and the Wagner Group. The operators wrote mainly in Russian and appeared to be based in Russia; because OpenAI does not allow access from Russia, the company said the operators used VPNs to reach ChatGPT.

The Iranian network of seven fake journalists

The Iranian operation worked through seven fictitious journalist identities that pitched articles to online publications. OpenAI found nearly 100 articles published or republished under those bylines across about a dozen outlets, mostly small and midsize sites covering international affairs, geopolitics and the Middle East. The earliest appeared in July 2025 and the latest in October 2026, and OpenAI did not name the publications.

Operators used ChatGPT to refine long articles, write pitches to editors and draft internal reports, and they generated batches of English and Persian social media comments on political topics. According to The Record, those posts drew little engagement, although some host outlets had large audiences, and one had nearly 2 million Facebook followers. Iran International, the London-based Persian-language broadcaster, reported that the campaign also repeatedly generated hostile social media responses aimed at it.

Arise adds that output rose significantly after the U.S.-Iran conflict broke out, that one persona, Ervin B. Hoskins, posed as an American freelance writer with accounts on X and Instagram, and that the activity looked like a commercial influence-for-hire service. OpenAI could not identify the client and said it shared information with relevant authorities.

OpenAI’s assessment and its published threat reporting

OpenAI’s own verdict, quoted by The Record, was that the operations “closely resembled complex influence operations of the pre-AI age, but used AI to make some of the workflows easier.” Both networks went beyond fake social accounts and used seemingly independent organizations, a think tank in one case and a stable of bylined freelancers in the other, to carry their messages. The accounts the company found were closed after the report’s findings, and the reporting gives no count of how many were banned.

The company publishes these findings on its security news page and has done so since at least February 2025, when its report on disrupting malicious uses of AI named Ben Nimmo among its authors. Neither page listed the October 2026 report as of Oct. 10, so the specifics above are those carried by The Record, The Insight International and Arise, each of which attributes them to OpenAI.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.

More from Morning Overview