ChatGPT Just Caught a Spy Stealing From OpenAI

In late September 2024, an OpenAI employee found a small device wrapped in toilet paper plugged into the back of a television in an executive conference room. It looked like a hacked-together USB stick with unusual circuitry. When the security team examined it, they realized it was an audio listening device, designed to record conversations in a room where OpenAI executives and business partners discussed proprietary product plans. What followed was an internal investigation that used OpenAI’s own AI model to identify the suspect, trace his planning, and ultimately lead to his arrest—a case the FBI’s director, Kash Patel, called a top priority, marking AI as an official matter of national security.

The Device and the Investigation

The device had three notable features: heavy cryptography protecting its stored data, a trigger to start recording only when someone began speaking, and encrypted data transfer. This wasn’t a simple off-the-shelf bug—it was meticulously designed. OpenAI handed the device to the FBI but also ran their own analysis using their latest internal model (the host speculates it might be GPT-6.5 or GPT-7). The model reverse-engineered the circuitry and, combined with CCTV footage, identified a likely suspect: Abdul Rahman Al-Fosan, a 35-year-old Saudi national who had worked at OpenAI for over a year on the product team, including on AI agents.