OpenAI has banned a cluster of ChatGPT accounts it says originated in Russia, after discovering they were being used to promote a fabricated think tank called the International Burke Institute (IBI) — a self-described “expert community” claiming to be based in Israel. According to OpenAI’s report, the operation used AI-generated social media posts, misattributed academic work, and a “sovereignty index” designed to cast Russia in a favourable light while criticising Western governments.
The takedown, reported by CNBC on 25 August, is the latest in a series of influence operations OpenAI has disrupted since the start of the war in Ukraine. What sets this one apart is not the audience it reached — which was small — but the infrastructure behind it.
How the operation worked
The operators prompted ChatGPT in Russian, instructing it to generate English-language social media comments for Substack, Telegram, X, Facebook, and LinkedIn. They explicitly told the model to hide any linguistic clues pointing to Russian origins. Since OpenAI does not allow access from Russia, the operators used VPNs to reach the platform.
Most of the generated content promoted the IBI website, which was registered in February 2025. The site listed prominent academics — including Francis Fukuyama and Noam Chomsky — as associated experts. OpenAI’s review of 36 articles published on the site found that 34 were copied from elsewhere on the internet, some years old, some attributed to the wrong authors. One article on the China-Pakistan Economic Corridor appeared to be copied from a Cambridge University Press original but was attributed to a University of Nottingham professor whose expertise is in a different field.
Alongside the think tank content, one operator generated German-language posts for a Telegram channel called “Lahme Ente” (lame duck) that routinely criticised Ukraine, the EU, and the German government. A second operator generated logos for a dozen Telegram channels focused on Germany, the USA, France, Poland, and Türkiye.
The sovereignty index
The operation’s most distinctive feature was a “sovereignty index” — a proprietary-looking ranking system that praised Russia while denigrating Western countries. Sample reports cited by OpenAI include descriptions of France as a “safe with historical capital” being “sold off in parts,” and Germany needing “politicians whose first loyalty is to the German people, not to corporate boards of directors in New York.”
This is the first time OpenAI has disrupted a Russia-linked operation that went to such elaborate lengths to manufacture institutional credibility. The index, the copied academic work, and the fake expert roster combined to create what OpenAI described as “an otherwise credible-appearing institution.”
Limited reach, outsized concern
The operation’s actual impact was modest. Social media posts received low engagement, and IBI accounts had small subscriber counts. Telegram channels attracted more attention — generally 10,000 to 20,000 followers each. Using the Brookings Breakout Scale, OpenAI assessed the operation at the lower end of Category Three: multiple platforms, with some indications of breakout to authentic audiences.
But the significance, as OpenAI noted, lies in the infrastructure rather than the audience. The operation built a fake institution with purported experts, republished academic content, and a proprietary risk index — assets that could be scaled over time. “This illustrates how influence actors can use AI as a supporting tool within a broader effort to manufacture authority, obscure the source of favored narratives, and establish assets that could be scaled over time,” OpenAI said.
CNBC contacted the Russian embassy in London and Russia’s Foreign Ministry for comment. No response had been reported at time of writing.
The pattern OpenAI keeps finding
This is not the first such takedown. In February, OpenAI shut down ChatGPT accounts connected to Rybar, a pro-Russia media company the UK government describes as “partially co-ordinated by the Presidential Administration.” In May 2024, OpenAI reported disrupting five covert influence operations that used its models. The pattern is consistent: AI is not the weapon itself, but a supporting tool that lowers the cost of producing plausible-looking content at scale.
What stands out here is the gap between effort and result. The operators built a fake think tank, copied dozens of academic papers, created a proprietary index, and maintained accounts across at least five platforms. They used ChatGPT to generate promotional posts. The content reached relatively few people. But the infrastructure remains — and as OpenAI noted, the supporting use of AI is also what led to the broader operation being exposed.
Why this matters beyond Russia
This connects to a broader question we have tracked at Singularity.Kiwi: the gap between AI’s capacity to generate convincing content and society’s capacity to verify it. The EU AI Act’s content transparency rules, enforceable from 2 August 2026, require AI-generated media to be marked in a machine-readable format. The C2PA content provenance standard — which OpenAI itself adopted in May — is part of that infrastructure.
But as security researcher David Buchanan demonstrated the same week, C2PA’s camera-based provenance is fundamentally breakable on Android devices. The technology layer that’s supposed to help distinguish real from fake has its own cracks. And New Zealand’s own deepfake legislation passed its first reading unanimously earlier this month — but criminal law alone doesn’t address the infrastructure problem.
The IBI operation shows both sides of the equation. AI made it cheaper to produce the promotional content. But it was the AI-generated posts that gave investigators the trail to follow. The same tool that enables the operation also exposes it.
📰 Sources
- OpenAI — Disrupting a new covert influence campaign from Russia
- CNBC — OpenAI bans Russian ChatGPT accounts used in covert misinformation campaign
- Brookings — The Breakout Scale: Measuring the impact of influence operations
- OpenAI — Disrupting deceptive uses of AI by covert influence operations (May 2024)
