OpenAI has exposed a covert Russian influence campaign that used ChatGPT to generate social media posts, and disrupted it by banning a cluster of accounts.

The operators accessed the platform from Russia through VPNs and instructed ChatGPT to hide any linguistic clues pointing to their Russian origin. The campaign's goal was to promote the "International Burke Institute," a think tank supposedly based in Israel. Among other things, it published a "sovereignty index" that ranked Russia favorably while making Western countries look worse by comparison.

Stolen expert identities

The most concrete part of the campaign is forgery. According to OpenAI, 34 of 36 expert-linked articles the institute published between September 2025 and May 2026 were copied from other sources, some with fake author credits.

Two examples show the method:

  • An article from Cambridge University Press was falsely attributed to a professor at the University of Nottingham.
  • A piece from the Migration Policy Institute was credited to an Australian food chemistry professor.

The second example also shows the campaign's carelessness: attributing a text on migration policy to a food chemist is a mismatch anyone checking sources would notice. But the campaign was not aimed at readers who check.

Low reach, scalable infrastructure

Posts spread across X, LinkedIn, Facebook, Substack and Telegram. Some content was in German, posted to a Telegram channel called "Lahme Ente" that criticized Ukraine, the EU and the German federal government while pushing for closer ties with Russia. A second operator created logos for about a dozen Telegram channels focused on Germany, the US, France, Poland and Turkey, and regularly requested Russian-language summaries of channel activity.

The results cut both ways. Individual posts got very few views and the institute's official accounts had low subscriber counts. But the linked Telegram channels each reached 10,000 to 20,000 followers.

OpenAI rates the campaign at category three of six on the Brookings Breakout Scale, meaning it spread across multiple platforms with early signs of reaching real users. The company says current reach stayed small, but the elaborate infrastructure behind it could have been scaled up over time.

A third case

This is not the first Russian operation OpenAI has caught on its own platform. In June 2024, the company exposed a network called "Bad Grammar" that used its models to generate political comments on Telegram about Russia, Ukraine and the Baltic states. A year later it identified "Operation Helgoland Bite," which produced German-language content ahead of Germany's 2025 federal election, attacking the US and NATO while promoting a political party.

Why it matters

What these cases share is that all of them were caught on OpenAI's own platform. The company can only look at its own system.

Given Chinese AI providers and freely downloadable open-weight models, the actual scale of AI-powered influence operations is likely much larger. An operator running an open-weight model on their own server appears in no provider's account records and has no account to ban.

So announcements like this describe a measurement problem as much as an enforcement success: what we see is limited to what gets caught on platforms that can be audited.