OpenAI announced Thursday that it has suspended accounts belonging to Russian and Iranian operatives who utilized its artificial intelligence tools to conduct influence operations across multiple nations. Internal documents generated via ChatGPT, which the company reviewed during its investigation, revealed that these actors successfully placed false narratives into established news outlets and even triggered responses from members of the European Parliament.
The company found that campaigns aiming to insert misleading stories into reputable media reached a broader audience than those relying solely on fabricated social media profiles. Russian operations, which primarily targeted Latin American countries, appeared designed to erode support for Ukraine and interfere in regional politics. According to OpenAI, the operatives mostly used ChatGPT to compile detailed work reports for their handlers, rather than to generate public-facing propaganda.
One particularly far-reaching operation involved Russian agents impersonating a regional educational authority in Peru. They deceived several schools into hosting events focused on Step an Bandera, a controversial historical figure admired by some Ukrainian nationalists but reviled in Poland and elsewhere for his movement’s collaboration with Nazis and persecution of Poles and Jews. The Peruvian outlet Extra confirmed that such an event occurred at a Lima school. Subsequently, a Polish news report on the incident prompted a far-right Polish member of the European Parliament to propose banning individuals who express support for Bandera.
“The Russian fake thus both fed on, and fed into, historical tensions between Ukrainians and Poles,” OpenAI stated in its analysis. Darren Linvill, co-director of Clemson University’s Watt Family Innovation Center Media Forensics Hub, noted that such activities are consistent with long-standing Russian strategies of “narrative laundering” and creating fake personas to advance political agendas.
Internal memos also claimed the Russians operated a fake think tank in Latin America, recruiting unwitting researchers to interview experts and draft reports. OpenAI compared this tactic to a 2020 operation linked to Yevgeny Prigozhin’s Internet Research Agency, which sought to influence the 2016 U.S. election. Linvill observed that while the operatives used AI to shortcut their workload, this reliance ultimately helped investigators trace their activities.
In contrast, Iranian operatives employed AI to produce content in both Persian and English, focusing heavily on the U.S.-Iran conflict and, to a lesser extent, American politics. OpenAI described the operation as resembling a commercial, for-hire influence campaign, though the company could not identify the specific actor behind it. Iranian accounts created seven fake author personas and published nearly 100 articles in small media outlets worldwide, including the U.S. site Daily Kos and the U.K.-based Middle East Monitor.
The Middle East Monitor has since removed the author pages for the fake contributors, while Daily Kos stated that its community section, where one persona posted, operates independently from staff reports and allows user-generated content without pre-publication review. Linvill emphasized that AI provided Iranian actors with more than just efficiency; it enabled them to produce language indistinguishable from authentic writing, a feat he noted would have been significantly more difficult just a few years ago.
This reminds me of the old IRA tactics, just faster. AI didn’t change the strategy, only the speed at which lies travel.
Wait, OpenAI is basically doing their own forensic investigation now? That’s a pretty big expansion of their responsibility.
Iranian operatives generating fluent English articles shows how dangerous accessible AI is. We need better media literacy urgently.
The idea that fake narratives in Peruvian schools could ripple through Polish politics is genuinely unsettling. This gets complicated fast.
So they used the AI to write reports, not propaganda? That seems like a clumsy mistake that helped investigators trace them.