OpenAI said in a Thursday blog post that it had banned Russian and Iranian accounts using its models to run covert influence operations across multiple countries. The Russian operation was assessed at Category 5 on the company’s IO Breakout Scale, the first operation to reach that level, and appeared to have co-opted unwitting people in Latin America to run an on-the-ground research front, including impersonating a Peruvian educational authority to stage Ukraine-themed school activities that later drew a response from a European Parliament member. The Iranian operation, rated Category 4, ran seven fake journalist personas pitching long-form articles to small and medium outlets worldwide and generated social media comments related to the US-Iran war.
OpenAI said the operations were notable for how closely they resembled complex pre-AI influence operations, such as the fake journalist Alice Donovan and the PeaceData front, while using AI to streamline workflows. The company said it has now exposed 30 covert influence operations over the past two and a half years, and found that campaigns landing content in established media outlets reached wider audiences than those relying on fake social media accounts.