OpenAI has blocked a cluster of ChatGPT accounts that it says very likely originated in Russia and were being used as part of a previously unreported covert influence operation. The campaign centered on the International Burke Institute (IBI), a self-described expert community that presented itself as being based in Israel while promoting material that portrayed Russia favorably and criticized Western countries.
According to OpenAI’s investigation, the operators used ChatGPT primarily to generate social media content promoting articles associated with the IBI. The posts appeared across platforms including X, LinkedIn, Facebook, Substack and Telegram. Many of the generated comments were written in English even though the operators prompted the system in Russian, and OpenAI said the users explicitly sought to remove linguistic clues that could reveal their Russian origin.
A sophisticated attempt to manufacture credibility
What makes the operation notable is not simply its use of generative AI, but the wider infrastructure built around it. OpenAI said the IBI website, registered in February 2025, presented itself as a credible research institution and claimed links to prominent experts. Yet a review by the company found extensive copying and misattribution of academic material. Of a sample of 36 articles linked to experts on the website and published between September 2025 and May 2026, OpenAI said 34 had been copied from elsewhere on the internet.
The campaign also promoted a so-called sovereignty index that presented Russia in a favorable light while ranking or characterizing Western countries less positively. OpenAI described the construction of this apparently authoritative institution, complete with purported experts, republished research and a proprietary-looking index, as unusually elaborate compared with other Russia-linked influence operations it has disrupted since the beginning of the war in Ukraine.
ChatGPT used as a supporting tool
The investigation illustrates how generative AI can serve as one component of a much larger influence operation. The actors did not rely on ChatGPT to create the core IBI articles identified by OpenAI. Instead, they used the model to produce promotional posts and comments designed to distribute and amplify material hosted elsewhere.
Some of the generated content was posted through accounts carrying IBI branding, while other posts appeared to come from ordinary users. The operators also generated replies to genuine Substack users, typically promoting the IBI channel. In another part of the network, ChatGPT was used to create German-language posts for a Telegram channel that criticized Ukraine, the European Union and the German government while advocating improved relations with Russia.
OpenAI also identified an operator who generated logos for multiple Telegram channels focused on countries including Germany, the United States, France, Poland and Türkiye. The same operator repeatedly requested Russian-language summaries of activity on those channels, adding another signal connecting the network to Russian-speaking operators.
VPNs and efforts to conceal origin
OpenAI said access to its models is not permitted from Russia. According to the company, the operators therefore used VPN services to reach the platform. Their prompts also included requests intended to conceal Russian linguistic characteristics in the resulting content.
These details are significant because they show that the campaign involved more than automated content generation. The actors appeared to be actively managing attribution, presentation and distribution, using AI to make material appear more natural to foreign audiences while attempting to obscure where the activity originated.
Limited reach, but a potentially scalable model
Despite the sophistication of the infrastructure, OpenAI assessed the campaign’s immediate impact as relatively limited. Typical social media posts attracted low numbers of views, while official IBI accounts had modest subscriber numbers. Some associated Telegram channels were more successful, with OpenAI reporting audiences generally in the range of 10,000 to 20,000 subscribers.
Using the Brookings Breakout Scale, OpenAI placed the operation at the lower end of Category Three, indicating activity across multiple platforms with some evidence of reaching authentic audiences. The company emphasized, however, that the more important issue was the infrastructure the operators had created and its potential to be expanded over time.
AI is becoming part of a broader influence toolkit
The case adds to a growing record of state-linked and politically motivated actors experimenting with generative AI. OpenAI has previously disrupted other Russia-origin influence activities that used its models to generate multilingual articles, comments and social media material. Similar investigations have also identified influence networks originating in other countries.
The latest operation suggests that the key risk may not always be mass production of obviously AI-generated propaganda. Instead, generative AI can be inserted into a broader ecosystem of websites, social accounts, fabricated institutional identities and repurposed legitimate material. In that environment, AI can help operators translate, localize and promote narratives while making coordinated activity appear more organic.
For platforms and researchers, the challenge is therefore increasingly about identifying the surrounding network rather than focusing only on individual pieces of generated text. OpenAI’s investigation shows that suspicious AI activity can also provide clues that expose a larger influence infrastructure.
A continuing battle over information integrity
The disruption does not mean that AI-driven influence campaigns have achieved large-scale success. In this case, OpenAI’s own assessment suggests that the campaign’s reach remained limited relative to its elaborate construction. But the operation demonstrates how inexpensive generative tools can support efforts to manufacture credibility and maintain multilingual influence networks across several platforms.
As AI systems become more capable and widely available, detecting abuse will increasingly depend on coordination between model providers, social platforms, researchers and other organizations monitoring deceptive online activity. The Russia-origin campaign uncovered by OpenAI offers a clear example of how generative AI can become one layer in a much broader attempt to shape public perception while concealing the identity and intentions of those behind it.