Russian accounts banned by OpenAI used ChatGPT to produce reports about influence operations in Latin America, including an effort that involved impersonating a regional education authority in Peru, the company said Thursday. The campaigns appeared intended to sway local politics and weaken support for Ukraine, according to OpenAI’s review of the accounts’ activity and materials they generated with the chatbot. In Peru, operatives claimed to have persuaded schools to organize activities about Ukraine that featured Stepan Bandera, a sharply divisive figure in the country’s history. Some Ukrainian nationalists regard Bandera as a leader in the fight for independence. Others, including many in Poland, condemn his movement’s alliance with Nazi Germany and its role in the killing of Poles and Jews. A Peruvian outlet reported that one such school event took place. A subsequent report in Poland prompted a far-right member of the European Parliament to propose barring people who express support for Bandera from entering the country, according to OpenAI. The company said the operation exploited and deepened historical tensions between Ukrainians and Poles. OpenAI said the Russian accounts also claimed to have operated a fake Latin American think tank and recruited researchers who did not know they were contributing to the effort. In some cases, the accounts took credit for events they had not caused. The company said the accounts used AI to create some fabricated material, but that most of their ChatGPT use appeared to involve writing updates for their superiors. Darren Linvill, a co-director of Clemson University’s Watt Family Innovation Center Media Forensics Hub, said the operation reflected a broader practice of using fabricated stories and identities to circulate narratives favorable to Russia. In this case, he said, the operatives’ use of AI to prepare reports helped expose their activities. OpenAI said the Russian efforts that sought coverage in established news outlets appeared to reach more people than campaigns focused on fake social media accounts. The company described the Peru operation as its widest-reaching influence effort identified so far, citing the public response from a European political figure. The company also said Iranian accounts used ChatGPT to create and refine Persian- and English-language content focused mainly on U.S.-Iran relations, and to a lesser extent on U.S. politics. OpenAI said it could not identify who was behind the effort, which appeared consistent with a commercial influence campaign. The Iranian accounts used seven fabricated author identities to pitch and polish long-form pieces, publishing nearly 100 articles in outlets around the world, OpenAI said. The outlets included the U.K.-based Middle East Monitor and Daily Kos’ community section. NBC News reported that Middle East Monitor published at least 21 articles by four of the fake authors; the outlet later removed the author pages. Daily Kos said its community posts are separate from staff reporting and are not reviewed by staff before publication. The posts attributed to the fake identity remained online. OpenAI’s findings highlight how generative AI can support influence campaigns beyond producing fake images or posts: it can help operators write convincingly across languages and provide a record of their work. Linvill said AI helped the Iranian effort produce writing that seemed authentic enough to be published by legitimate outlets.