OpenAI says Russia-origin accounts used ChatGPT to promote covert influence campaign

OpenAI says it has banned a cluster of ChatGPT accounts that very likely originated in Russia after tracing their role in promoting a covert influence operation built around a self described expert institute. The company published the findings on August 25, saying the accounts generated social posts for the International Burke Institute, or IBI, which presented itself as an Israel based think tank while pushing narratives favorable to Russia.

The important point for readers is not that a chatbot wrote propaganda posts. OpenAI says the campaign reached relatively small audiences. The more useful lesson is how ordinary AI output can sit inside a larger credibility machine. In this case, the company says the operation paired generated comments with a website that used copied academic articles, false or mismatched author attributions, an address in Israel, and a branded sovereignty index that praised Russia and criticized Western countries.

What OpenAI says it found

According to OpenAI, the operators used Russian language prompts to create social media comments for Substack, Telegram, X, Facebook and LinkedIn. Most of the generated posts were in English, and the users asked ChatGPT to hide linguistic clues that could reveal Russian origins. OpenAI also said the operators used VPNs because its models are not available from Russia.

The company reviewed a sample of 36 articles linked to experts on the IBI website and published between September 2025 and May 2026. It said 34 appeared to have been copied from other places on the internet. Some were years old, and others were attributed to people who did not match the original work. That detail matters because it shows the campaign was not only trying to generate volume. It was trying to borrow authority from real academics and institutions.

Why it matters

OpenAI rated the operation at the lower end of Category Three on the Brookings Breakout Scale, meaning it appeared across multiple platforms with some signs of reaching authentic audiences. The company also said several Telegram channels connected to the effort had roughly 10,000 to 20,000 followers each, even though typical posts drew low view counts.

For companies, campaigns like this are a warning about brand and source verification. A polished site, a formal sounding index, and posts that read like normal user comments are weak signals on their own. The practical rule is to check provenance before citing or sharing research, especially when a new institution publishes political rankings, security claims, or country comparisons without a clear record of authorship.

OpenAI’s report is also a reminder that detection can come from the supporting workflow, not only the public material. The same tools used to produce promotional posts can leave patterns that help expose the larger operation. That gives platforms a reason to monitor abuse signals around coordination, attribution, and account origin, while readers should treat expert branding as something to verify rather than trust at first glance.

Sources

• openai.com

Cover photo by cottonbro studio on Pexels, used under the Pexels License.