OpenAI says it has removed accounts tied to two unrelated influence campaigns, one Russian and one Iranian, that used ChatGPT as part of political propaganda efforts. The company laid out its findings in a safety report released on Thursday, as covered by The Record. Both operations pushed their messaging through channels that looked independent, such as a research center or bylined articles, rather than relying only on fake social media profiles.
Russian campaign: a research center that was a front
At the center of the Russian effort was the Social Research Center (SRC). It presented itself as a research platform examining the Indian diaspora in Latin America and India's ties with countries in the region. OpenAI says Russian operatives actually ran it behind an invented identity.
The campaign also hired people in Latin America to do research for the SRC, apparently without telling them who was behind it. OpenAI's assessment is that these workers acted in good faith and were unaware of the Russian link. The report says much of what the SRC published looked like those workers' own research, not machine-generated text.
According to OpenAI, the campaign's chief goal was to harm Ukraine's standing in Latin America. Some activity also seemed aimed at domestic politics in countries including Argentina and Bolivia.
ChatGPT's part, as the report describes it, was producing fake leaked documents and scripts for audio recordings. Some of that material circulated online. OpenAI says at least some of the false claims spread widely enough to trigger fact-checks and official denials.
Who was behind it
OpenAI did not pin the campaign on a named threat actor. It said the evidence points toward a connection with a Russian group called Politology, also known as La Compania. That group has been reported to be a successor to influence operations associated with the late Yevgeny Prigozhin and the Wagner Group.
The operators wrote to ChatGPT mostly in Russian and appeared to work from Russia. OpenAI blocks access to its models from the country, so the operators reached its services through VPNs.
Iranian campaign: seven invented reporters
The Iranian operation created seven fake journalist personas and used them to pitch online outlets on articles about the U.S.-Iran conflict. OpenAI identified close to 100 articles that ran, or were republished, under those names across about a dozen outlets. Most were small or midsize sites covering international affairs, geopolitics and the Middle East. OpenAI did not name them.
The report dates the earliest identified article to July 2025 and the most recent to October 2026.
ChatGPT helped the operators polish long-form pieces, write pitches to editors and draft internal reports. They also used it to generate batches of English and Persian social media comments on political and geopolitical subjects.
Those comments gained little traction, with few likes, views or replies. The publishing outlets were a different story: some had sizable audiences, and OpenAI says one had almost 2 million Facebook followers.
OpenAI says the accounts appeared to originate in Iran, mostly prompted the model in Persian and connected through VPNs. It did not name the people or organizations responsible.
What the two cases have in common
In both cases, the operators built or borrowed a veneer of independence: a think-tank-style center on one side, freelance bylines in real publications on the other. OpenAI summed up the pattern by saying the operations "closely resembled complex influence operations of the pre-AI age," with AI easing some of the work.
The reported uses of ChatGPT differed. The Russian operators used it to fabricate leaked documents and audio scripts, while the SRC's published research appears to have been largely human-written. The Iranian operators used it for editing, pitching, internal reporting and multilingual comment generation.
