How Russia Used ChatGPT to Build a Fake Think Tank

OpenAI has uncovered a covert Russian propaganda operation and banned the ChatGPT accounts behind it. At the center was the International Burke Institute, a supposed Israeli think tank that was in reality a Russian front. The case is a clear example of how generative AI lets actors build a seemingly credible infrastructure of studies, expert profiles, and social media presence with minimal effort, and how, in the end, it was an AI translation slip that helped blow the operation’s cover.

Key takeaways

• The International Burke Institute, registered in February 2025, presented itself as an independent Israeli think tank but was, according to OpenAI, run by Russian operators.
• Of 36 “expert” articles published between September 2025 and May 2026, 34 were plagiarized, some with falsely attributed authors, including a Cambridge University Press piece wrongly credited to a Nottingham professor.
• The campaign’s centerpiece was a self-invented “sovereignty index” that systematically rated Russia more favorably than Western countries, Russia’s military, for instance, scored half a point higher than that of the United States.
• Content spread across X, LinkedIn, Facebook, Substack, and Telegram; a German-language channel called “Lahme Ente” specifically criticized Ukraine, the EU, and the German government while calling for closer ties with Russia.
• OpenAI classified the operation as category three of six on the so-called Brookings scale: active across multiple platforms with early signs of genuine reach, but still limited overall impact.

How the front was built

The operators worked methodically. Since OpenAI blocks direct access from Russia, they logged in through VPN connections and worked mostly in Russian, then had ChatGPT rewrite their text so that linguistic traces of its origin disappeared. The International Burke Institute’s website then carried articles supposedly written by independent experts, but mostly copied from existing academic texts. In one case, a Cambridge University Press piece ended up credited to a British scholar in Nottingham who had never been involved in writing it. An article on migration policy was attributed to an Australian food chemist who clearly had nothing to do with the subject. From this backdrop, the institute regularly published its “sovereignty index,” a self-constructed ranking of countries by supposed independence that consistently favored Russia.

The giveaway hiding in the text itself

One detail in particular ended up revealing the campaign’s origin: several posts referred to Germany’s governing coalition not by its standard German name, “Ampelkoalition,” but as the “Svetofor coalition” — “svetofor” being the Russian word for traffic light. ChatGPT had apparently translated the term literally from the Russian-language source text without knowing or using the established German name, a small but telling slip that shows just how fragile linguistic cover can be even with AI-assisted translation. Traces like this are often exactly what analysts spot first when unmasking covert campaigns, long before technical metadata or IP addresses provide hard evidence.

How much damage was actually done

The immediate reach stayed modest. Individual posts and the International Burke Institute’s own accounts drew only a small number of views and followers. The impact was larger on the affiliated Telegram channels: the German-language “Lahme Ente” channel and other channels built by a second operator, focused on Germany, the United States, France, Poland, and Turkey, each reached 10,000 to 20,000 followers. That places the campaign in the middle of the Brookings scale OpenAI uses to classify such operations, not negligible, but far from a mass-scale disinformation effort either.

Why OpenAI is alarmed anyway

It is exactly this gap between limited current reach and elaborate infrastructure that OpenAI’s analysis emphasizes most. The real risk, the company argues, is not today’s click counts but the fact that a seemingly credible institute, complete with purported experts and its own professional-looking index, could be built within a few months and with comparatively little human effort. Such a front can be reactivated, expanded, or repurposed for new topics at any time. OpenAI describes the underlying pattern as generative AI helping actors simulate authority, obscure the true origin of a narrative, and build reusable infrastructure for future campaigns instead of starting from scratch each time. The case fits into a series: OpenAI points to earlier Russian operations such as “Bad Grammar” in June 2024 and “Operation Helgoland Bite” in 2025, which showed similar patterns and also ran through ChatGPT.

A pattern that keeps repeating

The International Burke Institute case is not an isolated one. When OpenAI previously exposed a fraud network based in Cambodia that abused ChatGPT to run automated investment scams, it observed a similar underlying pattern: criminal and state actors use the same tools legitimate users rely on to write text or do research in order to build convincingly real fronts, whether that means fake investment platforms or, in this case, a supposed think tank. What differs is rarely the technical tool, but the goal. For OpenAI, that means a constant balancing act between keeping a powerful tool openly accessible and catching its systematic misuse early, before a small front can grow into an influential propaganda machine.

What comes next

For readers, the case points to one main lesson: seemingly independent institutes, studies, and indexes now deserve closer scrutiny, especially when they push unusually one-sided geopolitical messages. A single plagiarized article or one linguistic slip like the “Svetofor coalition” is often enough to cast doubt on an entire front. OpenAI is unlikely to be the last to uncover a campaign like this, precisely because the tools involved keep getting cheaper and more accessible. How well platforms, media outlets, and individual users learn to spot these patterns on their own will likely determine how effective this kind of propaganda remains in the long run.