OpenAI blocked ChatGPT accounts linked to Russia for covert influence campaign
OpenAI has blocked a group of ChatGPT accounts that, according to the company's assessment, were highly likely operated from Russia and used as part of a previously undisclosed covert influence campaign. The operators used AI to promote a pseudo-expert project, social media posts, and the distribution of materials critical of Ukraine and Western countries.
OpenAI reported this on August 25 in a separate report on countering covert information operations.
The company found that users primarily interacted with ChatGPT in Russian, but requested content generation mainly in English. They also specifically instructed the model to remove linguistic features that could identify the Russian origin of the authors.
Since OpenAI does not provide access to its models from Russia, the operators connected to the service via VPN.
Created the appearance of an international expert center
A significant portion of the activity was related to the International Burke Institute (IBI), which presented itself as an expert community located in Israel. The project's website was registered in February 2025.
It featured materials supposedly by well-known experts and researchers. Among the people whose names appeared on the site were political scientist Francis Fukuyama and linguist Noam Chomsky.
However, OpenAI's verification showed that most of these publications were not original research by the institute. The company analyzed a sample of 36 articles published from September 2025 to May 2026 and found that 34 of them were copied from other sources.
In some cases, old academic works were published under other people's names. OpenAI clarifies that the articles on the IBI site themselves were not created using its models. ChatGPT was mainly used for their further promotion.
Materials were distributed across multiple platforms
Posts generated with ChatGPT appeared on X, Facebook, LinkedIn, Telegram, and Substack. Some messages were distributed by accounts with IBI branding, while others looked like pages of ordinary users whose main activity was promoting the project's materials.
The operators also created replies to posts by real Substack users. In such comments, they offered to subscribe to the IBI channel.
Separately, OpenAI identified a network of Telegram channels targeting audiences in Germany, the United States, France, Poland, and Turkey. For some of them, AI was used to create names, logos, and content.
One German-language channel called Lahme Ente regularly published materials critical of Ukraine, the European Union, and the German government, while simultaneously advocating for improved relations with Russia.
Another channel presented itself as an American news outlet. However, the operators managing the corresponding ChatGPT accounts regularly requested summaries in Russian about the operation of these platforms.
Created a favorable “sovereignty index” for Russia
One of the central elements of the project was the so-called sovereignty index. IBI presented it as its own system for assessing the independence of states.
According to OpenAI, the methodology was designed to show Russia in a favorable light while criticizing Western countries. Some materials were dedicated to France, Germany, the United States, and the European Union, including states actively opposing the Russian invasion of Ukraine.
OpenAI calls this campaign new and previously publicly undescribed. The company notes that unlike many other information operations, a rather sophisticated infrastructure was created here, designed to imitate the work of a full-fledged analytical center — with experts, research, and its own index.
Campaign audience remained limited
Despite the scale of the infrastructure created, OpenAI estimates the direct impact of the operation as relatively small. Most social media posts received few views, and official IBI accounts had a small audience.
More visible were Telegram channels associated with the operation — some of them had about 10 to 20 thousand subscribers.
Using the Brookings Breakout Scale for information operations, the company classified the campaign at the lower boundary of the third category. This means that the content was distributed across multiple platforms and in some cases reached a real audience, but did not yet achieve wide reach.
OpenAI considers the main risk not the current popularity of the project, but the infrastructure created. Such a scheme allows operators to gradually build the appearance of an independent expert source, conceal the origin of the theses being spread, and scale the information campaign in the future.
