On Thursday, OpenAI announced that it had disabled a set of accounts originating from Russia and Iran after discovering that the users were employing the company’s chatbot, ChatGPT to support coordinated influence operations. The shutdown affected profiles that were actively generating false narratives for distribution across a variety of media outlets in several regions, including Latin America, Europe and the United States.
Russian influence campaign in Latin America
The investigation revealed that Russian operatives focused on Latin American audiences, crafting narratives designed to erode support for Ukraine and to sway local political dynamics. According to internal logs, the actors used ChatGPT to draft progress reports for unnamed superiors, describing how they had infiltrated established news sources. One especially notable scheme involved impersonating a regional education authority in Peru convincing schools to host activities that highlighted Stepan Bandera. Bandera’s legacy is deeply polarising: while hailed by some Ukrainian nationalists, he is condemned in Poland for wartime collaboration with Nazi forces.
A Polish news outlet covered the Peruvian school event, prompting a far-right member of the European Parliament to call for bans on individuals expressing support for Bandera. OpenAI summed up the episode, noting that “the Russian fake thus both fed on, and fed into, historical tensions between Ukrainians and Poles.”
Iranian disinformation network
Separate probes uncovered a parallel effort emanating from Iran. Seven fabricated author profiles were used to produce and polish lengthy pieces that were then pitched to small-scale media sites worldwide. In total, the operation generated almost 100 articles with at least 21 pieces appearing in the U.K.–based Middle East Monitor and additional submissions to the U.S. progressive outlet Daily Kos. The Middle East Monitor later removed the author pages after the discovery.
The Iranian campaign targeted narratives surrounding the ongoing U.S.–Iran conflict and, to a lesser degree, broader American politics. OpenAI’s analysis likened the activity to a “for-hire” influence service, though the party commissioning the work remained unidentified.
OpenAI’s response and expert commentary
OpenAI concluded that influence attempts focusing on placement within established media achieved broader reach than those relying chiefly on fabricated social-media personas. The company also highlighted a similarity between the recent Russian effort and a 2020 operation linked to the late oligarch Yevgeny Prigozhin’s Internet Research Agency which attempted to sway the 2016 U.S. election.
Commenting on the findings, Darren Linvill co-director of Clemson University’s Watt Family Innovation Center Media Forensics Hub, explained: “Russians are constantly engaging in these sorts of active measures, trying to launder narratives, create fake stories, and craft personas that align with their worldview.” He added that the use of AI offered a “shortcut” that ultimately helped investigators trace the activity.
While the majority of the workload involved conventional research and outreach, the integration of ChatGPT allowed the actors to produce language that “looked and felt authentic,” facilitating publication in legitimate outlets. OpenAI’s crackdown underscores the growing challenge of policing AI-enhanced disinformation across borders.
