On 8 October 2026, OpenAI disclosed in a safety report that it had banned two covert influence operations originating in Russia and Iran. The company said the operators used ChatGPT alongside other methods to support false-front organisations and place geopolitical messaging before audiences. It rated the Russian campaign Category 5 on its six-level IO Breakout Scale, the first operation it has disrupted at that level, and the Iranian campaign Category 4.
Key points
- OpenAI banned accounts associated with two operations it traced to Russia and Iran.
- The Russian operation appears to have directed a research platform staffed by people in Latin America who were unaware they were working for a Russian group.
- The Iranian operation used seven fabricated journalist personas to pitch articles to online publishers.
- OpenAI assessed the operations at Categories 5 and 4, respectively, on a scale of potential reach and impact.
Dark Clark and the Social Research Center
OpenAI called the Russia-origin operation Dark Clark, after a persona named Mia Clark through which the operators appeared to control the Social Research Center, or SRC, in Latin America. It said the banned ChatGPT accounts were used for work targeting countries across the region. Much of the activity appeared intended to damage Ukraine’s standing, while some appeared aimed at domestic politics, particularly in Argentina and Bolivia. The operators connected through VPNs because OpenAI does not permit access to its models from Russia.
According to OpenAI, the operators’ principal use of ChatGPT was drafting and revising reports for an unknown superior. Those reports covered efforts concerning Ukraine and recruitment for its armed forces, political activity in Latin America and the management of the SRC. The operators also used the model to create campaign material and reports on the platform’s performance. OpenAI said their accounts sometimes sought credit for activity in which they had played no part.
The SRC was more than a website in OpenAI’s account of the campaign. Reports written by the operators discussed pay, recruitment and dismissals, which the company said indicated control of the organisation rather than cooperation with it. The available evidence indicated that its employees in Latin America did not know they were working for a Russian group. OpenAI said most of the SRC’s material appeared to have been produced originally by those staff, and called it the most complex front identity it has disrupted in two and a half years.
The operators’ reports claimed to have circulated false stories across the region. OpenAI said some matched fakes that open-source researchers have attributed to an entity known as Politology or La Compania. The company also identified open-source evidence of fact checks and official denials prompted by material associated with Dark Clark. It has shared information about the case with relevant authorities.
Bogus Bylines placed articles under invented names
The Iran-origin operation, which OpenAI named Bogus Bylines, worked through seven fabricated journalist identities. OpenAI said they pitched long-form pieces to small and medium-sized online outlets internationally. Almost 100 articles were published or syndicated under those personas, Unite.AI reported. Both operations placed content in established media outlets, according to OpenAI, although it said that some of the material was not generated with its models.
The Iranian operation also generated batches of social-media comments, generally about the US-Iran war, OpenAI said. They drafted internal accounts of their work with the model, using methods the company described as questionable or deceitful to overstate their effectiveness. The Iranian operators used VPNs to obscure their location, Unite.AI reported.
OpenAI compared the fabricated journalists with Alice Donovan, an identity it described as a front for Russian military intelligence whose writing appeared in Western outlets in 2016 and 2017. It also pointed to PeaceData, a fake news outlet run in 2020 by people associated with earlier Russian influence activity, which recruited journalists unaware of who was behind it. OpenAI said both earlier fronts ceased operating after exposure.
OpenAI’s Category 5 assessment rests on reach
OpenAI’s IO Breakout Scale runs from Category 1, the lowest, to Category 6, the highest, and assesses potential reach and impact. Across the 30 covert influence operations the company says it has exposed over two and a half years, it has found that campaigns placing material in outside publications tend to have greater potential reach than those relying on fake social-media distribution. Dark Clark’s Category 5 assessment is OpenAI’s first for an operation it has disrupted.
The distinction matters for these cases because publication gave the operators an audience beyond accounts they controlled. OpenAI said artificial intelligence can make the production of such material easier and help operators present it in fluent language. Its account of the SRC also describes decisions about staff and an organisation producing its own work — activities that depended on people as well as generated text.
OpenAI said one false story that Dark Clark’s operators claimed to have planted in Peru reached the point of contributing to public tensions between Ukraine and Poland.
_-1.jpg){kind=link}