OpenAI said it shut down a Russian operation, known as Dark Clark, and an Iranian operation, both of which used ChatGPT to create fake personas and front entities.
In May 2026 a false email reached schools in a Lima district asking for events on Ukraine that cited Stepan Bandera, and some schools complied and sent photographs.
The Iranian operation used seven profiles posing as Western journalists, and OpenAI said the articles ran from July 2025 to October 2026.
OpenAI referred to roughly a dozen outlets, while the Washington Post reported at least 20 news outlets for about 100 AI-generated articles.
OpenAI said on Thursday that it had shut down two influence operations, one run from Russia and one from Iran, that used ChatGPT to create fake journalist personas and to support a purported think tank spreading anti-Ukraine propaganda. The Russian campaign is known as "Dark Clark", a name taken from "Mia Clark", a fictitious identity used to manage the fake research centre. OpenAI said it suspended the accounts involved and described the Russian operation as the one with the highest impact score among those it has disrupted since 2024. OpenAI is led by Sam Altman and is based in California.
The Russian operators used ChatGPT mainly to write periodic reports for an unnamed superior, describing the activities they carried out and the results they obtained. They wrote in Russian and connected through VPNs, because access to ChatGPT models is not permitted from Russia. The operation had three fronts: denigrating Ukraine and hindering recruitment into its army, interfering in politics in Bolivia and Argentina, and running a fake think tank. That centre, the Social Research Center (SRC), presented itself as a research platform on Latin America. OpenAI said the operation co-opted "unwitting people" in the region, and the recruited researchers did not know the real purpose of their work.
In May 2026 an email reached schools in a district of Lima, apparently signed by a regional education directorate. It asked schools to organise events about Ukraine for the national day of cultural and linguistic diversity on 21 May, and it named Stepan Bandera, the Ukrainian nationalist who is a hero of independence to some Ukrainians but is associated in Russia and Poland with fascism, collaboration and atrocities. Some schools complied, confirmed the events and sent photographs. The address was false, and so was the whole episode. Peruvian and Polish media then reported the events, accompanied by the accusation that Kyiv was "exporting" ultranationalist ideologies.
One report said a Polish MEP proposed declaring persona non grata people who show "anti-Polish sentiments", while another said the proposal targeted support for Bandera. OpenAI said the episode had an effect beyond Latin America.
“Thus, the falsehood created by the Russians exploited the historical tensions between Ukrainians and Poles and, at the same time, fuelled them.”
The Iranian operation used seven profiles posing as journalists from Western countries, including the United States, to place long-form articles in media outlets around the world. Its articles ran from July 2025 to October 2026, and OpenAI said their frequency increased after the US-Iran war broke out earlier this year. One persona, Ervin B. Hoskins, described himself as "an American freelance writer" and had accounts on X and Instagram. X's transparency information linked his account to a "West Asia android app", and Instagram's placed it in Iran, according to screenshots OpenAI provided. A piece under his byline in Beirut Today portrayed the United States as an unreliable mediator between Lebanon and Israel. Both of Hoskins's accounts appeared to be suspended.
“We identified almost 100 articles published or syndicated under the operation's bylines across roughly a dozen online outlets around the world.”
The reach figures differ between reports. OpenAI's own statement, as quoted by France 24, refers to roughly a dozen outlets, while the Washington Post reported that about 100 AI-generated articles were published across at least 20 news outlets. OpenAI said the articles appeared mostly in small and medium outlets focused on international affairs, geopolitics and events in the Middle East. The Operation also generated social media comments on the US-Iran war. The Washington Post reported that the Hoskins byline had also appeared on articles critical of the war in Daily Kos and Middle East Monitor.
OpenAI said it had suspended the accounts behind both operations.
Pollar is free and carries no ads in the text. Its members keep it going.
Support PollarConflicts · Lima and San Francisco
The Russian operation used ChatGPT to write internal reports and run a fake Latin American think tank, while the Iranian one planted almost 100 articles under fake Western journalist bylines.
AI text based on reports by Il Fatto Quotidiano, Washington Post, lastampa.it and other publishers. Published .