Analysis
OpenAI banned a cluster of ChatGPT accounts originating in Russia that were being used to support a covert online influence campaign, CNBC reported, discovered after the company investigated a set of AI-generated social media posts that led to what it described as a much broader operation. Pulse previously covered OpenAI's broader safety and enforcement track record as the company has faced repeated scrutiny over how it detects and discloses this kind of misuse. Working in Russian, the operators used ChatGPT to draft social media content that appeared on Substack, Telegram, X, Facebook and LinkedIn, explicitly directing the model to strip out any features of the text that might reveal its AI origin.
The campaign centered on a fabricated organization called the International Burke Institute, a self-described expert community whose website -- registered in February 2025 and listing a street address in Israel -- falsely claimed prominent academics including Francis Fukuyama and Noam Chomsky as affiliated experts. OpenAI reviewed 36 articles linked to experts listed on the site, published between September 2025 and May 2026, and found 34 of them had been copied from elsewhere on the internet, in some cases with incorrect attribution to the fictional experts.
- OpenAI -- identified and banned the accounts, published findings in a public writeup
- International Burke Institute -- the fabricated think tank at the center of the campaign, falsely claiming real academics as affiliates
- Substack, Telegram, X, Facebook, LinkedIn -- the platforms where AI-generated content from the campaign appeared
- Francis Fukuyama, Noam Chomsky -- real academics falsely claimed as experts affiliated with the fake institute
Operators routed their ChatGPT access through VPNs specifically to circumvent OpenAI's existing restriction on model access from Russia, a workaround that highlights a persistent limitation of geography-based access controls: a determined, resourced actor can route around them relatively easily, meaning the restriction functions more as friction than a hard barrier for a state-linked operation.
OpenAI's own framing is worth taking seriously as the real point of the disclosure: the company said the operation's significance lies less in the audience it actually reached -- which appeared limited -- than in the infrastructure it had built. A fabricated think tank with a real-looking website, plausible-sounding fake experts, and years of planned content cadence represents a reusable template that could be redeployed or scaled by other actors, which is a more durable concern than any single campaign's immediate reach.
The counterweight is that OpenAI catching and disclosing this campaign, including the specific mechanics of how it evaded access restrictions, is itself evidence the detection and moderation systems are working as intended in at least this case -- the story is as much about a platform's response capability as it is about the underlying threat. Whether other AI labs are catching comparable operations at a similar rate, or whether this is simply the one that happened to surface publicly, remains an open question none of the major labs have addressed with comparable transparency.
