OpenAI has banned a “cluster” of ChatGPT accounts that it says formed part of a pro-Russia influence operation designed to disparage Western countries while promoting a think tank built on misattributed academic research.

In a blog post this week, the company said the banned users employed VPNs to circumvent its restriction on Russian users, then used its technology to create social media posts promoting the International Burke Institute (IBI) — an organization that describes itself as an Israel-based “expert community” but which OpenAI identifies as the “core” of a Russian influence operation.

According to OpenAI, the think tank — whose website remains online — lists prominent figures such as American linguist and intellectual Noam Chomsky and American political scientist and author Francis Fukuyama among its contributors. Neither Chomsky nor Fukuyama responded immediately to Fortune’s requests for comment. One author listed on the site said she had no idea the website was using her name until Fortune reached out to her. Other listed experts, including Shlomo Avineri and Jiang Ping, died before the IBI was founded.

OpenAI said the website, registered in February 2025, also hosted a “sovereignty index” that favored Russia, alongside dozens of articles carrying misattributed authorship. A Fortune review of the Burke Institute’s website via the Wayback Machine found that although the site existed as of February last year, nothing was published on the page before October 2025.

The institute has pushed back against the allegations. Responding to reporting by French outlet Le Monde on OpenAI’s claims, it issued a statement rejecting any suggestion that it is not a legitimate think tank.

“The fact that third parties use or cite an IBI publication, the Institute’s name, or findings from the Sovereignty Index does not, in itself, constitute evidence of cooperation, coordination, funding, or a common underlying purpose,” the statement read.

The Burke Institute did not immediately respond to a Fortune request for comment sent to the email listed publicly on its website. OpenAI said that, according to its research, a handful of individuals in Israel represented the Burke Institute in interviews and at conferences.

Using real researchers’ names

OpenAI’s review of the Burke Institute’s articles found that 34 of 36 were genuine academic papers misattributed to other authors. A screenshot published by OpenAI showed a legitimate article on migration governance that appeared on the Burke Institute’s website: it had been taken from the Migration Policy Institute but was credited to Kate Howell, a professor of food science at the University of Melbourne, alongside a headshot pulled from the internet. Since OpenAI published its blog post, the website has changed the article’s attribution to one of its real authors, Kate Hooper, a senior policy analyst at the Migration Policy Institute. Hooper did not immediately respond to a request for comment.

Another professor listed as an expert on the Burke Institute’s website is Susan Ariel Aaronson, a professor of international affairs at George Washington University. A legitimate article she authored for the Centre for International Governance Innovation was posted on the website with a link to download.

Aaronson — who is also a co-principal investigator for the NSF-NIST Institute for Trustworthy AI in Law \u0026 Society (TRAILS) — told Fortune she is not associated with the Burke Institute in any way. Reached by phone, she described the situation as disturbing.

“Trust is really important in the governance of AI, and here you have this really weird, you know, situation causing substantial distrust, and so it’s just really disturbing,” she said.

She also lamented that there is little regulation at the federal level to deal with such instances, arguing that it needs to be handled from the top.

“The Trump administration is doing nothing to effectively govern it, and is governing AI in an opaque manner. That is opacity does not build trust,” she added.

Other articles on the Burke Institute’s website appear to have been translated into English from a Slavic language, according to OpenAI, which cited clues such as the use of the word “svetofor,” meaning “traffic light” in several Slavic languages, including Russian.

Amplification across social platforms

OpenAI claims the ChatGPT users created numerous social media posts to promote the institute. Some of these came from accounts bearing the International Burke Institute’s name and branding, while others came from made-up individuals who promoted the institute on Facebook, LinkedIn, and X, as well as Substack and Telegram.

OpenAI said posts from individual accounts “only received low numbers of views” and that the institute’s official accounts had “low subscriber numbers.” On the instant messaging platform Telegram, however, the operators were able to attract a larger audience of between 10,000 and 20,000 followers per channel.

Despite what OpenAI characterizes as a small following, the company’s findings illustrate how Russia-linked actors have reportedly used AI to step up their propaganda campaigns. For OpenAI, the case also shows how generative tools can be used not just to produce text, but to help assemble a broader credibility strategy around fake experts, copied research, and coordinated amplification across platforms.

Earlier this year, OpenAI said it disrupted another operation it traced to Russia, which used ChatGPT to generate German-language political content ahead of Germany’s federal election. That campaign criticized the U.S. and NATO, distributed content through Telegram and X, and created AI-generated material supporting Germany’s far-right Alternative für Deutschland party.

Separately, Russian-speaking cybercriminals used SpaceX’s AI coding assistant Cursor to hack into a Belgian chemical company along with six other companies this year alone, according to a Reuters report citing data from cybersecurity company Gambit Security.

Source: Fortune · OpenAI threat-intelligence report (PDF)