Spying at OpenAI: an engineer arrested by the FBI, unmasked by in-house AI

Only the paranoid survive. Andy Grove made it Intel’s motto. However, at OpenAI, the threat came from within. The FBI arrested Abdulrahman Alfozan, in-house engineer since July 2025, on October 9. He is suspected of having slipped an encrypted audio recorder behind the television in a management meeting room in the San Francisco premises.

OpenAI claims that its own models made it possible to trace it back to it. After its agents accused of having saturated a UN site, OpenAI finally has a story where its AI plays the role of policeman.

A microphone hidden in toilet paper at OpenAI

The complaint summarized by the US Department of Justice details the scenario. Around September 30, OpenAI employees discovered a small box plugged into the USB port on the back of a television, wrapped in toilet paper. Impossible to read its encrypted content. The investigation dates its installation to September 28.

DIY James Bond style.

Surveillance cameras show Alfozan near the room several times before the discovery. However, he had neither access to this room nor invitation to its meetings. This room hosted confidential technical, commercial and strategic discussions.

On October 5, the engineer returned to the room and looked behind the television, where the first device was sleeping. The same day, he allegedly forgot a bag in another room. OpenAI security found a second recorder almost identical to the first, with his wallet and ID.

When the chatbot becomes a prosecution witness

Before installing the microphone, Alfozan allegedly queried an internal productivity tool about building addresses, meeting rooms, their capacities and their agendas. He even reportedly asked where management and external partners were meeting.

Investigators also point to a chatbot account opened in the name of his wife, which shared several identifiers with his. Between September 24 and 30, this account allegedly asked how to build a streaming audio recorder, add Wi-Fi to it, improve live listening, and encrypt its SD card. On September 27 and 28, the questions asked about the device's custom header and its MAC address. All network equipment has a unique one, like a license plate.

What you type into an AI leaves a trace that the editor keeps and can cross-check. The complaint does not specify which chatbot the suspect used. OpenAI did not further detail what its models analyzed to identify it. The company only says it used it to understand the device and trace its origin, before contacting the FBI.

Industrial espionage and national security at OpenAI

OpenAI reacted on Weights are the billions of learned parameters that form the heart of an AI. OpenAI fired him.

FBI Director Kash Patel believes that protecting American innovation in superintelligence is essential to the country's security. He makes it an absolute priority for counterespionage. Federal prosecutor Craig Missakian is calling it a matter of national security. The suspect is a 35-year-old Saudi national.

In early October, OpenAI had already fired three AI security researchers (Jasmine Wang, Tomek Korbak and Mikita Balesni) for violating its rules on sensitive information. Nothing officially links the two files.