OpenAI Bans Russian Accounts Behind Fake Think‑Tank Campaign

OpenAI announced that it has shut down a cluster of ChatGPT accounts that were traced back to Russia. The accounts were employed to generate and circulate social media posts that promoted the International Burke Institute (IBI), a self‑described Israeli think‑tank, and a “sovereignty index” that praised Russia while criticizing Western nations. The operation was uncovered after an investigation into AI‑generated comments on platforms such as X, LinkedIn, Facebook, Substack and Telegram.

The banned accounts were operated from Russia, a country from which OpenAI does not allow direct access to its models. The operators used VPNs to reach the platform and instructed ChatGPT to produce content in English while masking any Russian linguistic cues. The generated posts were posted by accounts that bore the IBI name and logo, as well as by seemingly ordinary users whose primary activity was to share IBI articles. In addition to promotional posts, the operators replied to real Substack users with requests to follow the IBI channel, further amplifying the institute’s reach.

The IBI website, registered in February 2025, presented itself as an expert community based in Israel. It claimed to host articles by world‑class scholars such as Francis Fukuyama and Noam Chomsky. A review of 36 articles linked to experts on the site revealed that 34 were copied from other sources, often with incorrect attribution. Some pieces appeared to be machine‑translated from Slavic languages, and the site also published a proprietary “sovereignty index” that compared countries on a scale that favored Russia. The index was not generated by OpenAI’s models but was promoted through the same AI‑generated social media content.

Compared to earlier Russia‑linked influence operations that OpenAI has disrupted, this campaign stands out for its elaborate construction. While previous cases involved isolated posts or small networks, this operation built a seemingly legitimate institution, complete with a website, a roster of purported experts, republished academic work, and a proprietary index. It is the first time OpenAI has encountered a Russia‑linked influence operation that combined AI‑generated promotional material with a fully fleshed‑out front‑end designed to appear authoritative.

The reach of the campaign was modest. Most social media posts received few views, and the official IBI accounts had low subscriber numbers. Telegram channels associated with the operation attracted between 10,000 and 20,000 followers each. Using the Brookings Breakout Scale, OpenAI assessed the operation as at the lower end of Category Three, indicating multiple platforms with some breakout to authentic audiences. Nevertheless, the infrastructure built by the actors, an AI‑generated content pipeline, a website with fabricated expertise, and a proprietary index, creates a foundation that could be scaled if the actors chose to.

OpenAI’s response underscores the importance of monitoring AI‑generated content for influence operations. The company’s detection tools were able to identify the Russian origin of the accounts, the use of VPNs, and the linguistic patterns that suggested an attempt to hide the operators’ nationality. The incident also highlights the challenges of distinguishing legitimate content from AI‑generated propaganda, especially when the content is repurposed from existing academic work.

Looking ahead, OpenAI plans to strengthen its moderation policies and to collaborate with other AI providers and social media platforms to share indicators of suspicious activity. The incident serves as a cautionary tale for developers, businesses, and policymakers about the potential for AI to be weaponized in covert political campaigns. It also raises questions about how to balance open access to AI models with the need to prevent their misuse in influence operations.

In sum, the disruption of this Russia‑linked campaign illustrates both the sophistication of modern influence operations and the critical role that AI providers can play in detecting and dismantling them. The event signals a need for continued vigilance and for the development of more robust tools to identify and mitigate AI‑enabled political manipulation.