OpenAI has banned two covert influence operations from ChatGPT and rated the Russian one Category 5 on its Breakout Scale for reach, the first Category 5 case it has disrupted since it began reporting takedowns.
That operation ran a fake research outfit, the Social Research Center, through an invented persona, aiming material at Latin American audiences to damage Ukraine’s reputation and to influence politics in Argentina and Bolivia. OpenAI says it posted more than 60 articles on its own site, fabricated leaked documents and audio, and appears to have drawn in local staff who did not know who they worked for. A second operation, attributed to Iran, used seven fake journalist personas to place roughly 100 articles across about a dozen small outlets.
The Breakout Scale tracks distribution, not persuasion, and OpenAI’s reading of its 30 reported cases is that operations climbed it by getting content into real media, while those confined to social platforms stayed near the bottom.
Both also used the models to write their own internal performance reports, and both overstated their results: one counted views on the posts it replied to rather than on its replies. These are OpenAI’s categorisations of its own enforcement data, with no outside audit, and the company notes the operations resembled pre-AI influence work with a faster drafting step.
Source: OpenAI, 8 October 2026.