OpenAI Uncovers Secret Scheme: ChatGPT Used as Disinformation Tool

OpenAI has shut down a group of ChatGPT accounts involved in a large-scale disinformation operation. The perpetrators used the company’s services to promote the “International Burke Institute” (IBI)—a pseudo-expert community allegedly based in Israel.

The Core of the Disinformation Operation

According to OpenAI’s internal investigation, the perpetrators attempted to lend legitimacy to the fictional “institute.” They used a website to publish academic papers concerning “sovereignty” in a manner favorable to Russia. Notably, the materials were often simply copied from real scientific works, sometimes with the addition of fake citations.

Mechanisms of ChatGPT Usage

Because official access to OpenAI’s neural networks is restricted in Russia, operators used VPNs to bypass these limitations. During their operations, they prompted the AI to generate comments for social networks such as Telegram, X (formerly Twitter), Facebook, LinkedIn, and the Substack platform. A key task for ChatGPT was masking linguistic markers that would otherwise reveal the authors’ Russian origin.

Investigation Details

According to the AI developer, the campaign was highly sophisticated, distinguishing it from previous attacks. Key manipulation tools included:

• Creating thematic social media posts on behalf of the “IBI institute.”
• Responding to comments from real Substack users to encourage subscriptions to the IBI channel.
• Impersonating ordinary users to legitimize the content being promoted.

Despite the complex structure, OpenAI assessed the audience reach as relatively small. It is worth noting that the IBI website was registered in February 2025. After analyzing the entire chain of actions, the group of Russia-affiliated accounts was promptly blocked to prevent further manipulation of public opinion via AI tools.

The IBI incident illustrates just how thin the line is becoming between real public dialogue and artificial imitation. With AI, it is possible not only to generate text quickly but also to manufacture an appearance of expertise, authenticity, and grassroots support. Although this specific campaign did not gain wide traction, the approach demonstrates a new reality in information warfare, where algorithms are no longer just tools for content creation but active participants in its dissemination and obfuscation.