OpenAI shuts down a Russian influence campaign that used ChatGPT to build a fake think tank

[Enterprise]

OpenAI announced this week that it has banned a cluster of ChatGPT accounts originating in Russia after uncovering a covert influence campaign built around a fabricated think tank called the International Burke Institute. The operators worked entirely in Russian, prompting ChatGPT to generate mostly English language content while specifically instructing it to strip out any linguistic features that might reveal the content's Russian origin, and they routed their connections through VPNs since OpenAI does not permit model access from Russia directly.

The fake institute presented itself as an independent Israel based expert community, complete with a website registered back in February 2025 that falsely claimed affiliations with well known real academics including Francis Fukuyama and Noam Chomsky. OpenAI's review of 36 articles published under the institute's name found that 34 had actually been copied from other sources, some with false authorship attached and others apparently machine translated from a Slavic language. The operation also produced what it called a sovereignty index, a ranking system that portrayed Russia favorably while criticizing the United States, France, Germany, and the European Union.

The campaign's actual reach stayed fairly limited according to OpenAI. Most of the social media posts generated across Substack, Telegram, X, Facebook, and LinkedIn received relatively few views, and the institute's own official accounts had small audiences, though a handful of affiliated Telegram channels focused on Germany, the US, France, Poland, and Turkey did pick up between 10,000 and 20,000 subscribers each. OpenAI was explicit that the significance of the operation wasn't about how many people it actually reached, it was about the underlying infrastructure the operators had built, an apparently credible institutional identity designed to manufacture authority and obscure the true source of the narratives it was pushing.

This isn't OpenAI's first disclosure of this kind. The company has previously disrupted campaigns tied to Chinese state linked actors and a pro Russia media outlet called Rybar, and it separately banned accounts earlier this year involved in campaigns targeting US debates over data centers. Taken together, these disclosures paint a picture of AI companies increasingly finding themselves acting as informal referees policing state linked information operations that route through their own products, a role none of them explicitly signed up for but one that's becoming a routine part of running a widely used consumer AI product

OpenAI bans Russian ChatGPT accounts used in covert misinformation campaignwww.cnbc.com

[TheGame]

This is part of a pattern that's plainly becoming routine at this point, every major lab periodically disclosing some state linked misuse of its own product feels like it's turning into its own little recurring news genre. Doesn't make it less important each time, just noting how normalized this specific category of disclosure has become in a fairly short window

[Bayley_US]

OpenAI framing the low view counts as almost beside the point is the correct read here. Building the infrastructure first and scaling reach later is exactly how a lot of successful influence operations historically play out, you don't need viral reach on day one if the actual goal is establishing a durable, reusable apparatus that can be activated more aggressively whenever it's actually needed

[Edward71]

The multilingual angle spanning German, French, Polish, and Turkish content alongside the main English push shows this was a really well resourced, deliberately structured operation rather than some small scale amateur attempt. Coordinating consistent messaging across that many languages and target audiences at once takes real planning and probably a decent sized team behind it, not just one person messing around with a chatbot in their spare time

[Highland Fatima]

The fake think tank angle is the part that actually worries me here more than the raw reach numbers. Fabricating an entire institutional identity, complete with claimed affiliations to truly respected real academics, is a fundamentally different kind of deception than just posting propaganda directly. It's laundering credibility through a fictional institution rather than just shouting an opinion into the void

[Runtime Arrow]

34 out of 36 articles being copied or misattributed is a pretty damning ratio once you actually sit with it. That's not a think tank doing original analysis with an AI assist, that's essentially a plagiarism operation wearing an academic costume, and the fact it took this long to get flagged says something uncomfortable about how thin the verification layer is for a lot of online expert commentary these days

[Dom9]

The VPN workaround for accessing a geo restricted model is such a mundane, boring detail buried in an otherwise fairly dramatic story, and it's a good reminder that geo blocking a model from a specific country is basically a speed bump rather than a real wall for anyone actually motivated to get around it. Doesn't mean the restriction is pointless, just that it was never going to stop a determined, resourced actor specifically

[Bleak Jason]

Curious how OpenAI actually detected this in the first place, since the writeup mentions it started from investigating AI generated social media posts more generally rather than looking specifically for this operation. Makes me wonder how much detection right now still depends on getting lucky with a broader sweep versus having a systematic way to catch this kind of thing reliably at scale

[Leopard96]

Wonder what actually happens to the underlying International Burke Institute website and its published content now that the accounts behind it are banned. Does the fake institute just quietly disappear, or does someone else eventually pick up the exact same shell and keep running it under new accounts once attention moves elsewhere