OpenAI has banned a cluster of ChatGPT accounts originating from Russia that were being used to run a covert influence operation, directing social media users towards a fake Israeli think tank sympathetic to Moscow. The company said the operation appeared to have had only limited reach, but its unusually elaborate setup made it stand out from previous Russia-linked campaigns OpenAI has disrupted, despite Russia officially being barred from accessing its services.

Posts and comments generated by the network appeared on Substack, Telegram, X, Facebook and LinkedIn, mostly written in English despite prompts to ChatGPT being submitted in Russian, with instructions to disguise their origin. OpenAI said the disguise attempts often failed, citing giveaway phrases such as an untranslated "Svetofor coalition" reference. Much of the content pointed to the "International Burke Institute", a supposedly Israel-based think tank that OpenAI found to be fabricated: of 36 sampled articles, 34 were copied from elsewhere, some misattributed to the wrong authors, including one migration policy paper wrongly credited to an Australian food science professor. The institute's "sovereignty index" consistently criticised Western nations at odds with Russia over Ukraine, though OpenAI said the campaign's social media posts generally attracted very low engagement.