OpenAI shuts down Russia-linked ChatGPT accounts over fake think tank

OpenAI has shut down a network of ChatGPT accounts tied to Russia.

The accounts used the chatbot to help create a fake “expert community,” including Francis Fukuyama and Noam Chomsky as contributors, the company said on Wednesday.

34 of 36 articles came from somewhere else

The accounts were there to promote the International Burke Institute (IBI). The outfit claimed to be based in Israel and staffed by well-known scholars.

OpenAI’s report says its website launched in February 2025. The roster it advertised by mid-July included Fukuyama and Chomsky and other big academic names.

OpenAI’s investigators probed a sample of 36 articles posted between September 2025 and May 2026 that were ascribed to so-called experts at IBI and found that 34 of them were taken from elsewhere.

There was a piece on the China-Pakistan Economic Corridor, which originated in a Cambridge University Press publication by Bradford University’s Yunas Samad, but IBI ran it under the byline of South Asian politics scholar Katharine Adeney, of Nottingham.

An article on migration governance in the Migration Policy Institute was published under the byline of an Australian professor of food science.

The account managers typed their prompts in Russian but requested output in English most of the time. OpenAI found that they told the model to remove any linguistic evidence that Russians were behind the text.

The company blocks access from within Russia, so operators used VPNs to access the service.

What ChatGPT produced was social media filler. Posts and replies sent readers to IBI articles across X, LinkedIn, Facebook, Substack, and Telegram.

Some were profiles that had the IBI name and logo on. Others posed as regular users whose feeds, when looked at, did nothing more than spew out IBI links.

The accounts also made responses to unrelated posts on Substack, each urging readers to follow the IBI channel.

One mistranslated German word broke the cover

OpenAI said the effort was more complex than the Russia-linked operations it has dismantled since the war in Ukraine started, despite its reach being small.

A “sovereignty index” praising Russia and criticizing Western governments was the cornerstone.

One article on Germany referred to “the Svetofor coalition,” using the Slavic word for traffic light, where a native writer would have said “traffic-light coalition.”

OpenAI took that to mean a machine translation word-for-word from a Slavic original, it said in the report.

Another German Telegram channel, “Lahme Ente” (lame duck), put out posts attacking Ukraine, the EU, and Berlin, calling for warmer ties with Moscow.

Some of the linked Telegram channels had as many as 20,000 followers each, but the individual posts saw little traffic.

In February, Cryptopolitan reported that a Chinese influence campaign gave itself away when an official used ChatGPT as a work diary.

Over and over, OpenAI has watched nation-state actors use the model for mundane tasks, not fresh attacks.

In 2024, it disrupted another Russia-linked network, Doppelganger, that generated posts praising Moscow and attacking Ukraine and NATO.

The smartest crypto minds already read our newsletter. Want in? Join them.