OpenAI said Tuesday that it has banned a cluster of ChatGPT accounts originating in Russia that were used to support a covert online influence campaign — an operation built around a fictitious research organization called the International Burke Institute that fabricated affiliations with prominent academics and plagiarized the vast majority of its published material.
The company disclosed the takedown in a blog post titled "Disrupting a new covert influence campaign from Russia," and the findings were reported by CNBC and Quartz. According to OpenAI, the accounts were used to promote the International Burke Institute, or IBI, a self-described "expert community" whose website listed a street address in Israel and was registered in February 2025. For more context on this story, see our ongoing more AI stories.
A Think Tank That Never Was
The operation's centerpiece was the appearance of institutional legitimacy. The IBI website falsely claimed to feature prominent academics including Francis Fukuyama and Noam Chomsky, OpenAI said. In a review of 36 articles linked to experts listed on the site and published between September 2025 and May 2026, the company found that 34 had been copied from elsewhere on the internet — sometimes with incorrect attribution.
A central feature of the campaign was a "sovereignty index" that ranked countries on a metric designed to cast Russia favorably while criticizing Western nations. Reports targeting France, Germany, the United States, and the European Union veered into polemic, OpenAI said. One entry on the United States read: "Trump arrived [in Beijing] with the No. 1 sovereignty index in the world and left after taking a number of steps that [...] can be qualified as a voluntary reduction in autonomy."
How the Operation Worked
OpenAI said it assessed the accounts as very likely originating in Russia despite the camouflage. To evade OpenAI's prohibition on model access from Russia, the operators routed their connections through VPNs, the company said. Working in Russian, they used ChatGPT to draft social media content that appeared on Substack, Telegram, X, Facebook, and LinkedIn — directing the model to strip out any features of the text that might betray its AI-generated origins. Most of the generated content was written in English, a choice that suggests the operation was aimed at Western audiences rather than domestic Russian readers.
The operation extended beyond English-language content. One operator generated German-language posts for a Telegram channel called "Lahme Ente" — German for "lame duck" — which criticized Ukraine, the European Union, and the German government while advocating closer ties with Russia. A second operator created logos for roughly a dozen Telegram channels focused on Germany, the United States, France, Poland, and Turkey, and requested Russian-language summaries of their activity.
Limited Reach, but an Alarming Blueprint
By conventional engagement metrics, the campaign was not a success. OpenAI acknowledged that most of the social media posts received low view counts, the official IBI accounts had few subscribers, and the associated Telegram channels attracted between 10,000 and 20,000 followers each.
But the company argued that the significance of the operation lies not in its reach but in the apparatus its operators constructed. Posts produced with ChatGPT steered readers toward what appeared to be a legitimate research body — complete with a physical address, an expert roster, and a steady stream of publications. That structure, OpenAI said, demonstrates how AI tools can help bad actors manufacture institutional credibility, obscure the true source of preferred narratives, and lay the groundwork for operations that could grow considerably larger.
The Broader Battle Over AI-Generated Influence
The takedown is the latest in a series of disruptions OpenAI has announced involving state-affiliated threat actors, and it arrives amid growing evidence that influence operations are shifting their ambitions from social media feeds toward AI systems themselves. Earlier in August, a separate investigation by Responsible Statecraft documented a fake think tank publishing more than 100 reports in a week in an apparent effort to manipulate the answers AI chatbots give about Israel and Palestine.
Together, the cases sketch out an emerging playbook: rather than only pushing messages to humans, operators are building false institutions whose output can be cited, scraped, and absorbed by the AI models that increasingly mediate how people encounter information.
For AI companies, the implication is uncomfortable. The same properties that make large language models useful — fluent multilingual drafting, low cost, tireless output — make them attractive infrastructure for laundered propaganda. OpenAI's response in this case was a content-level takedown: banned accounts, named tactics, and a public accounting of how the operation worked.
Whether such disclosures can keep pace with the operations they describe remains one of the central security questions of the AI era. As Tuesday's report makes clear, the cost of building a convincing fake institution has fallen dramatically — and the entities best positioned to detect them are the very companies whose models are being used to build them.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
