Analysis
OpenAI disclosed on August 25 that it banned a cluster of Russia-linked ChatGPT accounts used to run a covert influence campaign built around a fabricated think tank, the International Burke Institute, Fortune reported. The fake institute claimed affiliation with real, prominent academics -- including Francis Fukuyama and Noam Chomsky -- who were not actually involved.
OpenAI's own review found the operation's output was largely unoriginal: 34 of 36 IBI articles the company examined had been lifted from other sources online, then repackaged under the institute's fabricated academic branding. Operators used Russian-language prompts to generate English-language social content, then distributed it across X, Facebook, LinkedIn, Telegram and Substack -- a multi-platform footprint typical of coordinated influence operations rather than a single bad actor experimenting alone.
The operation took specific steps to evade detection: instructing ChatGPT to strip out linguistic patterns that might reveal Russian origin, and routing access through VPNs since ChatGPT is officially unavailable inside Russia. That's meaningfully more sophisticated than earlier waves of AI-assisted influence operations, which OpenAI and Meta have both disclosed periodically since 2023, typically catching cruder, more easily fingerprinted content.
“The Burke Institute case is a concrete, disclosed example of exactly the kind of AI-assisted operation that letter warns about, not a hypothetical.”
Some of the operation's Telegram channels carried real audiences -- 10,000 to 20,000 followers -- meaning the fabricated think tank had genuine, if modest, distribution before OpenAI's ban cut it off. That distribution scale sits well below what would move a national political conversation on its own, but it's large enough to seed narratives that get amplified further by real accounts sharing the content without knowing its origin.
This disclosure lands the same week OpenAI, Anthropic, Google and more than 110 other companies signed a joint letter warning that AI-enabled attacks -- cyber and influence alike -- will grow more sophisticated in coming months. The Burke Institute case is a concrete, disclosed example of exactly the kind of AI-assisted operation that letter warns about, not a hypothetical.
For platforms and AI labs, the pattern here is a genuine detection win: OpenAI caught and disclosed the operation rather than staying silent, which is the behavior policymakers have been pushing AI labs toward. The harder question is how many similar operations aren't being caught, given that this one only surfaced because of behavior patterns unusual enough to trigger review -- better-executed campaigns that don't share IBI's sloppy content-copying habit may not be getting flagged at all.
What to watch is whether other AI labs -- Google, Anthropic, xAI -- disclose comparable takedowns on a similar cadence, which would suggest this is now standard operating practice rather than an isolated response to one campaign.
