OpenAI said it banned two influence operations, one Russian and one Iranian, that used its AI models to run false fronts. A false front is an invented person or group that hides who is behind a message. The Iranian one used seven fake journalists to pitch long articles to online outlets. The Russian one, nicknamed Dark Clark, ran a Latin American think tank through a fake persona. OpenAI rated the Russian case Category 5 on a six-point reach scale, its first, and the Iranian one Category 4.
OpenAI said the Russian operators mostly used ChatGPT to write internal reports on their own work. Less often, they asked it to match accent and style by country, translate, and write audio scripts. They used VPNs, because OpenAI blocks access from Russia. The think tank's staff in Latin America wrote most of its original content and appear not to have known who ran it. OpenAI said some open-source evidence suggests some of the Russian fakes drew fact checks and official denials. It said both groups used deceitful methods to overstate their impact.
OpenAI said both operations closely resembled pre-AI ones, with AI easing some tasks. The weak point is still people: staff, editors and readers who trust a name. For teams that hire, publish or buy research, checks on who is hired, who pitches and who funds matter.