OpenAI has uncovered a covert Russian influence campaign that leveraged ChatGPT to generate social media content promoting pro-Kremlin narratives across the West. The campaign, which operated through Russian-based accounts using VPNs, aimed to bolster the credibility of the 'International Burke Institute' (IBI), a fake think tank purportedly based in Israel. The IBI published a 'sovereignty index' that ranked Russia's military higher than the U.S. military, according to the source text. | Image: Screenshot

The campaign spread content across platforms including X, LinkedIn, Facebook, Substack, and Telegram, with some posts in German targeting a Telegram channel called 'Lahme Ente' ('Lame Duck'). This channel criticized Ukraine, the EU, and the German government while advocating closer ties with Russia. Another operator created logos for about a dozen Telegram channels focused on Germany, the U.S., France, Poland, and Turkey, and requested Russian-language summaries of channel activity. A LinkedIn post from the campaign promoted an article by the IBI about Guyana's oil reserves, likely used as a hook to make the fake think tank appear credible for geopolitical analysis.

According to OpenAI, individual posts received very few views, and the official IBI accounts had low engagement. However, the linked Telegram channels each reached 10,000 to 20,000 followers. On the Brookings Breakout Scale, the campaign was rated category three out of six, indicating it spread across multiple platforms with early signs of reaching real users. OpenAI noted the campaign's current reach was small, but the elaborate infrastructure behind it could have been scaled up over time. OpenAI has previously disrupted several Russian influence operations abusing ChatGPT, including 'Bad Grammar' in June 2024 and 'Operation Helgoland Bite' in 2025. These are only the cases OpenAI caught on its own platform. With Chinese AI providers and open-weight models freely available, the actual scale of AI-powered influence operations is likely much larger.

Source: thedecoder