OpenAI Shuts Down Russian and Iranian Accounts Using AI for Disinformation Operations
OpenAI announced Thursday that it has terminated Russian and Iranian accounts that were exploiting its artificial intelligence tools to conduct influence operations across multiple countries. The campaigns involved operatives using ChatGPT to generate reports for supervisors and create content aimed at spreading disinformation, with some false narratives successfully reaching established media outlets and prompting responses from political figures.
The company's investigation revealed that influence campaigns focused on infiltrating traditional media achieved broader reach than those relying solely on fake social media profiles to disseminate their messaging.
Russian Operations Target Latin America
The Russian accounts primarily targeted Latin American countries with campaigns designed to undermine support for Ukraine and influence local political dynamics. According to OpenAI, the accounts were predominantly using ChatGPT to draft internal reports documenting their activities for higher-level operatives.
One Russian campaign represented what OpenAI described as the widest-reaching influence operation the company has encountered to date. In this operation, Russian operatives impersonated a regional educational authority in Peru and convinced schools to organize activities about Ukraine featuring Stepan Bandera, a controversial historical figure. Bandera led Ukrainian independence efforts before and during World War II but is condemned by many, particularly in Poland, because his movement collaborated with Nazi Germany and participated in violence against Poles and Jews.
The Peruvian outlet Extra confirmed that at least one such event occurred at a school in Peru. Polish media coverage of the incident prompted a far-right member of the European Parliament to propose banning entry to individuals expressing support for Bandera.
The Russian fake thus both fed on, and fed into, historical tensions between Ukrainians and Poles.
Darren Linvill, co-director of Clemson University's Watt Family Innovation Center Media Forensics Hub, noted that such activities align with established Russian tactics. He explained that Russia continuously engages in active measures involving narrative laundering and the creation of fake personas to generate or spread propaganda consistent with Russian political objectives.
The Russian internal documents generated with ChatGPT also claimed the operatives managed a fake think tank in Latin America and recruited unsuspecting researchers to interview experts and produce reports. OpenAI compared this effort to a 2020 Russian operation that posed as an independent news outlet, linked to late oligarch Yevgeny Prigozhin's Internet Research Agency, which played a role in attempts to influence the 2016 United States election.
In some instances, the Russian operatives took credit for events they did not actually orchestrate. While the Russians did employ AI to generate fabricated material, it constituted a minority of their overall workflow.
Iranian Campaign Targets Media Outlets
Unlike their Russian counterparts, Iranian operatives used AI primarily to assist in producing and disseminating content in both Persian and English. The campaign concentrated on the United States-Iran relationship and, to a lesser extent, American domestic politics. OpenAI assessed that the activity patterns suggested a commercial entity conducting influence operations for hire, though the company could not identify the specific actors involved.
The Iranian accounts created seven fictitious authors to pitch and refine lengthy articles for smaller media outlets worldwide. Through this method, they published nearly 100 articles across various platforms.
These fabricated authors successfully placed articles in outlets including the progressive American news site Daily Kos and the United Kingdom-based Middle East Monitor. NBC News reported that Middle East Monitor published at least 21 articles attributed to four fake authors. The outlet has since removed the author pages but did not respond to requests for comment.
Daily Kos stated in response to inquiries that its community section, where one fake persona posted content, operates separately from staff reporting. The platform allows anyone to register and post in the community section without prior review by Daily Kos staff, though posting rules prohibit users from pretending to be someone they are not. The posts from the fraudulent persona remain accessible on the site.
Linvill emphasized that AI provided value to the Iranian operation beyond simple efficiency gains. The technology enabled them to generate language that appeared authentic enough to pass editorial scrutiny at legitimate publications—something that would have been significantly more challenging for Iranian operations just a few years ago.
The disclosure highlights growing concerns about the potential for artificial intelligence tools to facilitate sophisticated disinformation campaigns and the challenges facing technology companies in preventing malicious use of their platforms.
