OpenAI Exposes Russian Influence Network Using ChatGPT to Fake a Think Tank
OpenAI has banned a cluster of Russian accounts that were using VPNs to dodge access restrictions and run a coordinated influence operation through ChatGPT. The accounts generated social media content distributed across Substack, Telegram, X, Facebook, and LinkedIn.
The central prop in the operation was something called the International Burke Institute, or IBI, presented as a global expert community based in Israel. Its website, registered in February 2025, claimed to bring together leading voices in economics, politics, sociology, and international relations. It did not.
What investigators actually found was a site stuffed with copied academic work, some of it falsely attributed to real scholars, alongside other content that appeared to have been written in a Slavic language and then fed through a machine translator. There is no direct evidence ChatGPT wrote the articles on the site, but the tool was clearly being used to produce the surrounding promotional noise.
The operation's most distinctive feature was a custom metric called the Sovereignty Index, also branded as the Burke Index. Calculated across seven dimensions including military, economic, technological, and cultural sovereignty, it was the kind of thing that sounds rigorous until you look at who it flatters. The United States topped the rankings with 650.9, China came second at 649.1, and Russia landed at a suspiciously respectable 601.4. The accompanying writeup described Russia as a state with 'record military and economic self-sufficiency' and 'full digital and military nuclear sovereignty.' Subtle, it was not.
OpenAI noted the campaign is distinct from previously documented Russian influence operations, mainly because of this manufactured index. The think tank framing was the point. A sovereignty ranking published by a credible-sounding institution carries more weight than a Telegram post saying Russia is great, even if both are equally hollow.
Audience reach was limited. Most Telegram channels attracted somewhere between 10,000 and 20,000 followers. The operators instructed ChatGPT to strip out any linguistic markers that might hint at Russian authorship, and the bulk of the output was in English.
One account used ChatGPT to generate a profile picture for a Telegram channel called American Observer and asked for Russian-language summaries of its content, suggesting the operators were monitoring their own messaging.
OpenAI's conclusion is worth quoting directly: the significance here is not the audience reached but the infrastructure built. ChatGPT was a supporting tool, used to generate a thin layer of promotional content pointing at an otherwise elaborately constructed fake institution. Academic credibility, fabricated experts, a proprietary risk index. The AI did not build the operation. It helped dress it up.
The operation also managed to expose itself precisely because it used AI. The generated posts left detectable traces that led investigators back to the wider network. Handy.