OpenAI bans ChatGPT accounts behind a Russian influence campaign
OpenAI says it banned a cluster of ChatGPT accounts that very likely originated in Russia after tracing them to a covert influence operation built around the International Burke Institute (IBI), a self-described "expert community" claiming a street address in Israel. The operators wrote their prompts in Russian but had ChatGPT generate mostly English social media comments, and they instructed the model to strip out any linguistic clues pointing to Russia. Because OpenAI blocks access from Russia, the operators reached the platform through VPNs. The generated posts were used to promote IBI content on Substack, Telegram, X, Facebook and LinkedIn, some from accounts carrying the IBI name and logo and others from apparently inauthentic "everyday user" accounts whose main activity was sharing IBI articles. On Substack, the operation also generated replies to real users' posts that pushed a request to follow the IBI channel. The IBI website itself, registered in February 2025, was not built with OpenAI's models. It presented itself as an expert network and, as of 17 July, listed prominent names including Francis Fukuyama and Noam Chomsky among its claimed roster. In a review of a sample of 36 articles on the site linked to named experts and published between September 2025 and May 2026, OpenAI found 34 had been copied from elsewhere on the internet, in some cases misattributed to the wrong author. One article on the China-Pakistan Economic Corridor appeared to be lifted from a Cambridge University Press original but was credited to a University of Nottingham professor whose actual expertise is South Asian politics. Another, on migration governance, appeared copied from the Migration Policy Institute but was attributed to an Australian professor of food science. Alongside the republished articles, IBI published its own "sovereignty index" reports on individual countries and comparisons between them, focused heavily on states critical of Russia's war in Ukraine, such as France, Germany and the European Union, plus similar material about the United States. OpenAI describes the tone of these reports as polemical, quoting lines such as one on France arguing the country "was opened like a safe with historical capital, and now it is being sold off in parts," and another on the EU titled "Slovakia vs Italy: The paradoxes of the Sovereignty of the Hostage-Countries of the 'Big' Union." Beyond IBI, one of the two identified operators also generated German-language posts for a Telegram channel called "Lahme Ente" (lame duck), which criticized Ukraine, the EU and the German government while calling for closer ties with Russia. A second operator generated logos for about a dozen Telegram channels focused on Germany, the United States, France, Poland and Turkiye, some of which also shared IBI content, and repeatedly asked ChatGPT for Russian-language summaries of what those channels had been posting. OpenAI says the operation's direct reach looks limited: its social media posts drew low view counts and the official IBI accounts had few subscribers, though its Telegram channels generally attracted 10,000 to 20,000 followers each. Using the Brookings Breakout Scale, OpenAI places the operation at the low end of Category Three, meaning activity across multiple platforms with some signs of reaching real, unaffiliated audiences. OpenAI also notes that some online traces suggest a handful of real individuals based in Israel may have represented IBI in article submissions and at conferences, but says it cannot determine what relationship, if any, those individuals have to the Russian operators. Its overall reading is that the case shows how AI tools can serve as a supporting piece within a larger effort to fabricate an authoritative-looking institution, and that this same reliance on ChatGPT for isolated tasks is part of what let the broader operation be traced and exposed.
Key facts
- OpenAI banned a cluster of ChatGPT accounts that very likely originated in Russia, used to generate social posts promoting the International Burke Institute (IBI), a self-described expert community claiming to be based in Israel.
- Operators prompted in Russian but had ChatGPT write mostly English posts with Russian linguistic clues stripped out, reaching the platform via VPNs since OpenAI blocks direct access from Russia.
- Of 36 IBI-website articles reviewed (published between September 2025 and May 2026), 34 were copied from elsewhere, including one Cambridge University Press piece misattributed to a University of Nottingham South Asian politics professor and one Migration Policy Institute piece misattributed to an Australian food science professor.
- A second operator generated logos for about a dozen Telegram channels covering Germany, the US, France, Poland and Turkiye, including a German-language channel, 'Lahme Ente,' that criticized Ukraine, the EU and Berlin.
- OpenAI judges the operation's reach as limited overall, with Telegram channels drawing 10,000 to 20,000 followers each, and rates it at the low end of Category Three on the Brookings Breakout Scale.
Why it matters
OpenAI frames this as the most elaborate Russia-linked influence operation it has disrupted since the start of the war in Ukraine: rather than just flooding platforms with AI-written comments, the operators built an entire fake institution, complete with a website, a roster of claimed experts, republished academic work and a proprietary-sounding 'sovereignty index,' and used ChatGPT only for the narrow task of driving traffic to it. That points to a pattern where AI is not the whole operation but a supporting tool that helps manufacture the appearance of institutional legitimacy.
Who it affects
The direct targets were users of Substack, Telegram, X, Facebook and LinkedIn who encountered IBI-branded accounts, comment replies, or the dozen Telegram channels focused on Germany, the US, France, Poland and Turkiye. The falsely credited academics at the University of Nottingham and in Australian food science had their names attached to republished work they did not write. OpenAI also flags that a handful of real individuals based in Israel may have represented IBI at conferences or in article submissions, though it says it cannot establish their relationship to the Russian operators.
How to use it
There is no product here, but the case offers a concrete checklist for spotting similar operations: a recently registered site presenting itself as an expert institute, articles attributed to real academics whose actual specialty does not match the topic, phrasing that reads like a literal machine translation from a Slavic language, and social accounts whose entire posting history consists of promoting one outside website. OpenAI's own detection process started from exactly this kind of anomaly in AI-generated social posts.
How solid is it
This is OpenAI's own investigation and self-report, not independent verification, but it is backed by specifics: a sample review of 36 IBI articles with a stated 34-copied count, named platforms, dated website registration, and direct quotes from the sovereignty index reports. OpenAI also applies an external yardstick, the Brookings Breakout Scale, to size the operation's reach rather than only describing it qualitatively.
Risks and caveats
OpenAI is explicit about what it does not know: it cannot determine the relationship between the real individuals possibly linked to IBI in Israel and the Russian operators, and it does not give a total count of banned accounts or an exact ban date, only 'recently.' Because the report comes from the company that also sells the model being misused, its framing of both the threat and OpenAI's own effectiveness at catching it is inherently self-interested.
“Ten years of Macronism produced exactly the result that a manager who came to the Elysee Palace from the Rothschild Bank must have given: France was opened like a safe with historical capital, and now it is being sold off in parts.”
— IBI 'sovereignty index' report on France, quoted by OpenAI