OpenAI said on Thursday it shut down two covert influence operations that used ChatGPT to run false-front journalists and think tanks, the first time the company has rated a campaign at Category 5 on its six-point breakout scale since it began publishing threat reports in 2024.
What OpenAI says it found
In a report titled “Disrupting AI-enabled ‘false front’ operations,” the company said it banned account clusters linked to Russia and Iran that blended traditional influence tradecraft with large language models. The Russia-origin campaign, which OpenAI calls “Dark Clark,” targeted Latin America with messaging aimed at undermining Ukraine’s reputation and, in some cases, shaping local politics in Argentina and Bolivia, according to the report. Operators prompted ChatGPT mostly in Russian, with some Spanish-language activity, and shared findings with authorities, OpenAI said.
The Iran-origin campaign, nicknamed “Bogus Bylines,” used seven fake journalist personas to pitch articles on the U.S.–Iran conflict to small and medium online outlets, OpenAI said. The company identified nearly 100 articles published or syndicated under those bylines across roughly a dozen sites between July 2025 and October 2026, with output accelerating after the latest Middle East fighting. OpenAI rated that operation Category 4 for article placement and Category 2 for social engagement, noting that many posts drew little reaction.
Why the company’s scale matters
OpenAI’s breakout scale is meant to measure how far an influence operation escapes niche channels and lands in mainstream conversation. Category 5, in the company’s framing, indicates content that triggered fact checks, official denials or political responses—signals that a narrative broke into public debate. Dark Clark’s operators also ran a fake think-tank website with more than 60 mostly original articles and, according to OpenAI, appeared to involve unwitting staff in Latin America who did not know they were working for a front organization.
Security researchers who reviewed the disclosure said the case illustrates a familiar pattern—state-aligned messaging laundered through respectable institutions—now accelerated by AI drafting, translation and pitch emails. “The operations closely resembled complex influence campaigns of the pre-AI age, but used AI to make some of the workflows easier,” OpenAI wrote.
What happens next
OpenAI said it has now publicly attributed 30 covert influence operations in roughly two and a half years of reporting. The company did not name a specific government agency behind either cluster, and Iranian activity was described as resembling a commercial for-hire campaign. U.S. officials have separately warned that foreign actors are experimenting with generative tools ahead of elections; OpenAI’s ban applies to its own platform, not to the outlets that published the material.
Editors and researchers said the disclosure shifts the question from whether AI can write propaganda to whether newsrooms can trace who is pitching them. Several outlets that carried Bogus Bylines pieces have not yet commented publicly. OpenAI said it will continue sharing technical indicators with platforms and governments, but noted that account bans alone cannot remove articles already indexed on the open web.