OpenAI's September 30 sweep found a hidden recorder

Federal prosecutors say an OpenAI engineer plugged an audio recorder into a USB port on the back of a television in an executive conference room at the company’s San Francisco headquarters, concealed it in toilet paper, and left it running. OpenAI’s technical surveillance countermeasures team found it on September 30 during a sweep. The company says the device went in around September 28, that its storage was encrypted, and that it carried audio monitoring and a data-transfer mechanism built to avoid monitoring.

The engineer, Abdulrahman Alfozan, 35, a Saudi national, has worked at OpenAI since July 2025 on the product team; a LinkedIn profile lists him as an engineering lead on the AI agents platform. He was arrested Friday and charged with using or attempting to use a device to intercept oral communications, and remains in custody ahead of a detention hearing on October 19. If convicted he faces up to five years and a $250,000 fine. He was not authorized to enter the room or attend meetings there, prosecutors say, though surveillance footage places him near it on several dates before the device turned up. Meetings in that room covered proprietary research, product plans, and government engagements.

The evidence trail in the complaint is the part worth studying. Before the install, prosecutors allege, Alfozan used a workplace productivity tool to ask about OpenAI’s building addresses, meeting rooms, room capacities and calendars, and where leadership and outside partners were meeting. That is reconnaissance, and it ran through an internal system that logged it. Separately, a chatbot account tied to his spouse, sharing identifiers with his own account, submitted prompts between September 24 and 30 about building, installing, obscuring and remotely accessing a recording device, including how to “build a continuous audio recorder” and “encrypt the SD card.” On September 27 and 28 that account asked about the specific device later recovered, down to its MAC address.

What put his identity on a second device was a bag. On October 5 he entered the conference room and appeared to look behind the television, then left a bag unattended in a different room. Security opened it and found his wallet and ID alongside a second recorder similar to the first.

OpenAI’s handling of the takedown is the useful operational example. Rather than fire him on discovery and tip off the investigation, it coordinated timing with the FBI. Between September 30 and the arrest, the company gradually restricted his access to systems and spaces, then suspended his physical and network access, placed him on leave, and secured his company-issued equipment. The firing came the day of the arrest.

For anyone running security at a company with rooms where sensitive conversations happen, the specifics map onto plain controls. Conference-room AV gear is an attack surface; a live USB port behind a television ran a recorder for roughly two days before a sweep caught it, which argues for periodic TSCM sweeps of sensitive rooms rather than against them. The reconnaissance was visible in advance to anyone correlating internal-tool queries against role and need to know, since an agents-platform engineer pulling leadership calendars and room capacities is a pattern worth flagging. And the phased, law-enforcement-coordinated revocation OpenAI describes is how you keep an insider from destroying evidence once they sense they are caught.

OpenAI says it used its own models to work out what the device was and to analyze forensic images of its storage, and the US attorney, Craig Missakian, called the case a matter of national security. The company says it has no indication that model weights, source code, or customer data were improperly accessed or taken, and no sign that product safety or customer information was compromised.