OpenAI said it banned Russian and Iranian networks that used ChatGPT to support fake journalists and a front think tank, with one placing almost 100 articles in news outlets.
Key Points:
- OpenAI banned two clusters of ChatGPT accounts from Russia and Iran that backed false front entities.
- Seven fake journalist personas placed almost 100 articles across roughly a dozen outlets, the company said.
- The Russian campaign was the first OpenAI rated Category 5 on the six-point Breakout Scale.
OpenAI ChatGPT Bans
The company disclosed the bans in a report published Thursday, describing two covert campaigns built around what it called false front entities. Both clusters of accounts reached ChatGPT through VPNs. OpenAI did not tie either campaign to a specific government agency, and said the Iranian activity resembled a commercial, for-hire operation.
The Iranian operation, nicknamed Bogus Bylines, ran seven invented reporters who posed as Western journalists and pitched long-form pieces on the U.S.-Iran conflict to editors. OpenAI identified almost 100 articles published or syndicated under those names across roughly a dozen small and medium outlets, the earliest dated July 2025 and the latest this month.
The Russian operation, called Dark Clark, targeted Latin America and worked mainly to damage Ukraine's reputation there. Its operators appeared to control a self-styled research platform, the Social Research Center, through a fake persona named Mia Clark. Available evidence indicates the center's local staff did not know they worked for a Russian group, the company said.
Also Read: Crypto ETF Momentum Returns After $50B Flows, JPMorgan Says
Breakout Scale Ratings
OpenAI rated the Russian campaign Category 5 on the six-point Breakout Scale, the first operation to reach that level since the company began publishing such reports. It cited evidence that the fakes led to public comment by politicians in several countries.
The Iranian article-planting effort earned a Category 4, while its social media comments rated a 2. Those replies drew likes and views in the single or double digits. Across the 30 covert operations it has exposed in two and a half years, OpenAI said those that land content in real outlets tend to have the widest reach.
Some Dark Clark fakes spread far enough to draw fact checks and official denials, including a fabricated audio clip claiming Bolivia's government would cut water to La Paz. The company cautioned that the operators also took credit for events they had no part in, so their own claims of impact cannot be taken at face value.
Earlier cases OpenAI disclosed gained far less traction. In August 2024, the company banned accounts tied to Storm-2035, an Iranian operation that generated content for five websites posing as news outlets, and rated it a Category 2.
OpenAI investigator Ben Nimmo said at the time that the operation tried to play both sides but did not appear to get engagement from either.
Read Next: OpenAI's 719 AI Math Proofs Under Fire: Was Navier-Stokes Lost In Translation?
