OpenAI uncovers Russian AI-driven disinformation network using ChatGPT

OpenAI has identified and disrupted a covert Russian influence operation that weaponized ChatGPT to produce and distribute pro-Kremlin narratives across Western digital platforms. The campaign, which operated from September 2025 to May 2026, relied on 36 ChatGPT accounts controlled by operators accessing the platform through Russian VPNs. These accounts were used to generate content for the fictitious International Burke Institute (IBI), a think tank purportedly based in Israel but with no verifiable presence. The IBI’s output included a so-called sovereignty index that ranked Russia favorably while portraying Western nations in a negative light, alongside other politically charged materials.

The campaign’s content was disseminated across multiple platforms, including X, LinkedIn, Facebook, Substack, and Telegram. Notably, some posts were published in German on a Telegram channel named "Lahme Ente" ("Lame Duck"), which criticized Ukraine, the EU, and the German government while advocating for closer ties with Russia. A second operator within the network created logos for nearly a dozen Telegram channels targeting Germany, the United States, France, Poland, and Turkey, often requesting Russian-language summaries of channel activity to maintain operational cohesion. While individual posts garnered minimal engagement, the linked Telegram channels collectively amassed between 10,000 and 20,000 followers each, indicating a modest but structured audience.

OpenAI’s investigation revealed that 34 of the 36 IBI articles published during the campaign were plagiarized from other sources, with some articles falsely attributed to academics who had no involvement. For example, an article originally from Cambridge University Press was credited to a professor at the University of Nottingham, and a piece from the Migration Policy Institute was falsely attributed to an Australian food chemistry professor. This pattern of misattribution suggests an attempt to lend false credibility to the IBI’s output by associating it with reputable institutions or experts.

The campaign’s reach, while currently limited, was structured to scale. OpenAI rated the operation at category three on the Brookings Breakout Scale, which measures the potential for influence campaigns to spread across platforms and reach real users. The infrastructure behind the campaign, including the use of AI-generated content, coordinated channel branding, and cross-platform distribution, demonstrates a sophisticated approach to digital disinformation, even if its immediate impact was constrained by low engagement metrics.

This is not the first time OpenAI has uncovered Russian influence operations abusing its models. In June 2024, the company disrupted "Bad Grammar," a network that used ChatGPT to generate political comments on Telegram about Russia, Ukraine, and the Baltic States. A year later, OpenAI identified "Operation Helgoland Bite," another Russian campaign that leveraged ChatGPT to produce German-language content ahead of Germany’s 2025 federal election. These campaigns targeted the US and NATO while promoting the AfD party, highlighting a pattern of AI-driven disinformation aimed at destabilizing Western institutions and public discourse.

The broader implications of AI-powered influence operations extend beyond OpenAI’s platform. With Chinese AI providers and open-weight models freely available, the actual scale of such campaigns is likely far greater than what has been detected. The use of AI to automate content creation and dissemination lowers the barrier to entry for state-sponsored disinformation, enabling actors to produce large volumes of convincing material at minimal cost. This raises concerns about the resilience of digital platforms and the effectiveness of current detection mechanisms in identifying and mitigating such threats.

The discovery also underscores the challenges faced by AI developers in balancing innovation with responsibility. While ChatGPT and similar models offer powerful tools for communication and creativity, they can also be repurposed for malicious purposes. OpenAI’s proactive disruption of these campaigns reflects an ongoing effort to identify and shut down abuse, but the cat-and-mouse nature of such operations means that new tactics will likely emerge. The company’s transparency in reporting these incidents provides valuable insight into the evolving landscape of AI-driven disinformation.

For policymakers, the revelations highlight the need for stronger safeguards and international cooperation to address AI-enabled influence operations. The Brookings Breakout Scale, used by OpenAI to categorize the campaign, offers a framework for assessing the threat level of such operations, but broader regulatory and technological solutions will be necessary to curb their spread. As AI models become more advanced, the potential for misuse grows, making it imperative for governments, tech companies, and civil society to collaborate on solutions.

Looking ahead, the focus will likely shift to improving detection methods and developing AI systems that can resist manipulation. OpenAI’s actions demonstrate the importance of vigilance and proactive measures in countering AI-powered disinformation. However, the broader challenge remains: as long as AI tools are accessible, state and non-state actors will continue to explore ways to exploit them for influence operations.