OpenAI says Russian group used ChatGPT to run fake think tank
OpenAI said Thursday it banned ChatGPT accounts tied to a Russian influence campaign in Latin America and a separate Iranian one that used fake reporters. The Russian group ran a research center whose local staff did not know who they worked for. OpenAI rated it a 5 on a 6-point scale, the first it has caught at that level.
OpenAI said Thursday it shut down two covert influence campaigns that used ChatGPT, one run from Russia and one from Iran. The company banned the accounts and said it shared what it found with authorities.
The Russian effort, which OpenAI calls "Dark Clark," aimed at Latin America. Its goals were to hurt Ukraine's image in the region and stir up local politics, especially in Argentina and Bolivia. The name comes from "Mia Clark," a fake persona the operators used to run a so-called think tank, the Social Research Center.
That center hired real people in Latin America to interview experts and write reports. "The available evidence indicates that the SRC's employees in Latin America were not aware that they were working for a Russian group," OpenAI said, according to AFP. The company called it the most complex fake front it has broken up in two and a half years.
The operators mostly used ChatGPT to write progress reports to their bosses, NPR reported. In those reports, they bragged that a fake email posing as a school office in Lima got Peruvian schools to hold Ukraine events that praised Stepan Bandera. Bandera was a Ukrainian nationalist. His movement killed Poles and Jews during World War II. A Polish report on the events led a far-right Polish member of the European Parliament to respond.
OpenAI warned that the operators also tried to take credit for events they had nothing to do with, CNA reported.
The Iranian campaign used seven fake "journalists" to pitch long articles about the U.S.-Iran war. OpenAI found almost 100 of them in about a dozen small and mid-sized outlets, dating from July 2025 to this month. NPR reported that one persona posted in the community section of Daily Kos, which anyone can join.
OpenAI rates these campaigns on a 1-to-6 scale. Most score a 1 or 2, CyberScoop reported. The Russian effort got a 5 and the Iranian one a 4, because their stories broke into real news coverage. OpenAI did not say which Russian group was behind Dark Clark, and it could not identify who ran the Iranian campaign.
What we know: OpenAI banned both networks and rated the Russian one the most far-reaching it has caught.
What we don't: Who paid for either campaign, and how many people actually saw the content.
Where outlets agree NPR, CyberScoop, AFP and CNA all report that OpenAI banned Russian and Iranian ChatGPT accounts and that the Iranian network used seven fake journalist personas.
Where they differ CyberScoop says the groups used ChatGPT to build fake personas and think tanks, while NPR reports the Russian operatives mostly used it to write progress reports to their bosses.