OpenAI said this week it had identified and shut down a covert Russian influence operation that used ChatGPT to manufacture the appearance of an independent foreign-policy think tank, part of a broader campaign to spread narratives favorable to the Kremlin across Western social media.
In a report published by the company, OpenAI said the operation centered on a fabricated organization calling itself the International Burke Institute, which presented itself online as an Israel-based "expert community." The group's website, registered in February 2025, listed prominent academics among its purported affiliates, though OpenAI said those names were used without basis. Of 36 articles the company reviewed that were attributed to the institute's listed experts, 34 were lifted from elsewhere on the internet, in some cases with incorrect attribution or machine-translated from Russian and other Slavic languages.
According to OpenAI, the operators worked mainly in Russian, using ChatGPT to draft material that was then published in English on Substack, Telegram, X, Facebook and LinkedIn. Because ChatGPT is not accessible from Russia, the operation's accounts connected through VPNs. OpenAI said it found the operators had explicitly instructed the model to strip out "linguistic features that could reveal that the content was produced by Russian speakers" — an attempt, the company said, to obscure the material's true origin.
A ratings gimmick and a small but growing audience
A recurring feature of the campaign was what OpenAI described as a "sovereignty index," a metric built to rank countries in ways that cast Russia favorably while criticizing the United States, France, Germany and the European Union. Most of the institute's own content drew limited engagement, and OpenAI characterized its official audience as small. Some affiliated Telegram channels, however, had accumulated between 10,000 and 20,000 subscribers, giving the network a real, if modest, distribution footprint beyond the institute's own site.
OpenAI said the case illustrates a pattern distinct from earlier propaganda efforts it has disrupted: rather than flooding platforms with volume, the operators invested in building an institution-shaped identity — complete with a name, a website and a stable of invented experts — that could lend borrowed credibility to Kremlin-aligned talking points over time. The company said this "attempt to construct an apparently credible institutional identity" was the more significant risk, even though the immediate reach was limited, because such infrastructure could be scaled up or reused for other campaigns.
The takedown follows OpenAI's disclosure in June that it had disrupted a separate, China-linked influence effort targeting U.S. debates over AI policy, part of a running cadence of threat reports the company has published as generative tools become a standard component of state-linked propaganda operations. Independent reporting on the Russian campaign by CNBC confirmed the broad outlines of OpenAI's account, though — as with prior such disclosures — the full scope of the network's real-world influence rests largely on OpenAI's own investigation rather than independent verification of engagement figures.
OpenAI said it banned the accounts associated with the operation and is sharing indicators with industry peers, a step it has taken in previous takedowns to help other platforms identify related activity.
