OpenAI bans accounts tied to Russian and Iranian influence operations

OpenAI said on 8 October that it had banned ChatGPT accounts tied to two covert influence operations originating in Russia and Iran. The operations used apparently independent journalists and a purported think tank to carry geopolitical messages to audiences through media outlets and online platforms, according to the company. The disclosure matters because OpenAI says some of the material reached published news sites, although the evidence does not show how many people believed it.

In its report on the operations, OpenAI rated the Russia-origin campaign Category 5 and the Iran-origin campaign Category 4 on its six-level Breakout Scale. It said the Russian case was the first Category 5 operation it had disrupted. Those ratings are OpenAI’s assessment of reach, rather than an independent measure of persuasion or political effect.

How fabricated Iranian bylines reached news outlets

OpenAI called the Iranian campaign ‘Bogus Bylines’. It said the operators created seven journalist identities and used ChatGPT to refine lengthy articles and pitches. The company identified almost 100 articles published or syndicated under those bylines across roughly a dozen online outlets. It described the outlets as generally small to medium-sized publications covering international affairs, geopolitics or events in the Middle East.

The earliest article OpenAI identified under the operation’s bylines appeared in July 2025 and the latest in October 2026. OpenAI said publication became more frequent after the US-Iran conflict began. Its account also describes prompts written in Persian, material produced in Persian and English, and the use of VPNs to obscure where operators accessed ChatGPT.

What OpenAI found in the Latin American campaign

The Russia-origin operation, which OpenAI named ‘Dark Clark’, focused on Latin American politics and culture and principally sought to damage Ukraine’s reputation in the region, according to CyberScoop’s account of the company’s findings. OpenAI said the operation used a purported Latin American think tank called the Social Research Center to put messages before audiences under a seemingly independent identity.

OpenAI said available evidence indicated that the center’s employees in Latin America did not know they were working for a Russian group. It described the center as producing a majority of its original content through those local staff. That distinction matters to the account of how the operation worked: the public-facing material did not depend solely on text generated by AI, while the apparent institutional identity concealed who was directing the activity.

The company said the Russian operation also made fabricated ‘leaked’ documents and audio scripts, some of which it found circulating online, and used AI extensively to draft internal reports. OpenAI called it the most complex attempt to run a front identity that it had disrupted in the previous two and a half years. CyberScoop reported examples of fact-checks and official denials prompted by material linked to the campaign, including claims circulated in Ecuador and Bolivia.

What the disruption does and does not establish

The two cases illustrate different routes into public discussion. The Iranian campaign put articles under invented individual bylines in existing outlets. The Russian campaign used an apparently independent organization and local staff to generate and circulate material. OpenAI said both blended AI assistance with methods used in influence operations before generative AI, rather than relying on automated content alone.

OpenAI compared the fake Iranian journalists with earlier fabricated media identities, including ‘Alice Donovan’, whose articles appeared in Western outlets in 2016 and 2017, and PeaceData, a fake news outlet that recruited unwitting writers in 2020. The comparison describes a recurring method: a credible-looking byline or organization can provide a route into established publishing channels.

The available accounts do not establish how many readers encountered or believed the planted articles, whether they changed anyone’s views, or what corrections individual outlets made. The attribution and breakout ratings rest on OpenAI’s investigation; CyberScoop reported the company’s findings but did not present a separate technical attribution. Those limits leave the scale of the campaigns’ real-world effects uncertain even though OpenAI says it has shut down the associated ChatGPT accounts.