OpenAI bans Russia-linked ChatGPT accounts promoting a think tank built on copied papers
The campaign used ChatGPT for posts and channel logos while an Israel-branded website republished academic work and praised Russia.
By Ryan Merket · Published
Primary source: OpenAI
Why it matters
The case shows generative AI functioning as the production layer for a larger credibility operation, supplying multilingual promotion and branding around copied research.
OpenAI banned a cluster of ChatGPT accounts that it said likely originated in Russia and supported a covert influence campaign spanning Telegram, X, Facebook, LinkedIn and Substack.
The operators used Russian-language prompts to produce mostly English social media comments promoting the International Burke Institute, a self-described expert community that claimed to be based in Israel, OpenAI said in a report published August 25th. They also instructed ChatGPT to remove linguistic clues that could reveal their Russian origins.
OpenAI said the users reached ChatGPT through virtual private networks because its models are unavailable in Russia. The report did not attribute the campaign to the Russian government. Its evidence links the ChatGPT accounts and their operators to Russia, while describing the International Burke Institute as the central public-facing asset of the operation.
AI handled the promotion
ChatGPT's main job was producing the material around the institute. The operators generated promotional posts, replies to real Substack users and German-language content for a Telegram channel called Lahme Ente, or "lame duck." That channel criticized Ukraine, the European Union and Germany's government while advocating closer ties with Russia, according to OpenAI.
A second operator used ChatGPT to create logos for about a dozen Telegram channels focused on Germany, the United States, France, Poland and Turkey. The same user repeatedly requested Russian-language summaries of the channels' activity. One US-focused channel posed as an American news outlet, complete with a generated profile image and awkward English in its biography.
The operators distributed content through accounts carrying the institute's name and branding as well as profiles that appeared to represent ordinary users. Those accounts largely posted links to the institute's material.
ChatGPT did not generate the academic articles or country reports hosted on the institute's website, OpenAI said. The model instead supplied repetitive production work that helped distribute and package the broader operation. That division of labor resembles earlier abuse cases documented by OpenAI, where threat actors used its models for translation, research, basic coding and content assistance inside larger, largely conventional operations.
A credibility shell built from copied research
The International Burke Institute website was registered in February 2025 and presented an Israeli street address. It listed prominent academics and published research alongside a proprietary "sovereignty index" that consistently portrayed Russia favorably and attacked Western governments.
OpenAI examined 36 articles associated with experts on the website and published from September 2025 through May 2026. It found that 34 had been copied from other sources. Some were years old, and several were assigned to the wrong authors.
One article concerning the China-Pakistan Economic Corridor appeared to reproduce a Cambridge University Press paper while crediting another academic. A migration article appeared to copy an April 2025 Migration Policy Institute brief, which names Meghan Benton, Natalia Banulescu-Bogdan and Kate Hooper as its authors, while the institute attributed it to an Australian food scientist.
The site also showed signs of machine translation. An article about Germany referred to its former traffic-light coalition as the "Svetofor coalition," using a transliteration of the Slavic word for traffic light. OpenAI assessed the phrasing as evidence that material had probably been translated from a Slavic-language original.
OpenAI said online traces indicated that several real people in Israel may have represented the International Burke Institute in conference appearances and article submissions. The report did not establish their relationship with either the institute or the Russia-based operators. That uncertainty makes OpenAI's attribution narrower than a claim that every person associated with the institute belonged to the same network.
The audience remained modest
Most promotional posts received few views, and the institute's official accounts had small subscriber counts, according to OpenAI. The associated Telegram channels showed greater reach, generally attracting between 10,000 and 20,000 subscribers each.
OpenAI placed the operation at the lower end of Category Three on the Brookings Breakout Scale. The six-level framework classifies Category Three campaigns as activity that crosses multiple social platforms and reaches multiple communities.
The operation's technical sophistication was limited. Its organizational design was more developed: copied scholarship supplied authority, an index supplied a repeatable message, and AI-generated posts supplied distribution across several languages and platforms.
That structure also created the trail that exposed it. OpenAI said its investigation began with AI-generated social posts and expanded into the institute, its copied research and the wider group of channels. ChatGPT gave the operators a faster production system, while their activity inside the model gave OpenAI visibility into a campaign that extended far beyond its service.