The FBI has arrested an OpenAI engineer on charges that he planted a hidden audio recorder in an executive conference room at the company's San Francisco offices—and the company says its own AI models helped identify the suspect. OpenAI says it has no indication that he took model weights, source code, or customer data.

Abdulrahman Alfozan, a 35-year-old Saudi national who has worked as an engineer at OpenAI since July 2025, was taken into federal custody on Oct. 9, according to a Justice Department announcement. He is charged with using or attempting to use a device to intercept oral communications and made his initial appearance in federal court on Tuesday. Prosecutors say a chatbot account tied to his spouse asked how to build, hide, and remotely access a continuous audio recorder in the days before the device was installed.

OpenAI said it discovered the device hidden inside one of its offices and turned to its own AI systems to trace where it came from.

"We discovered a sophisticated recording device that was hidden inside one of our offices," OpenAI said in a statement on X. "We used our models to understand the device and where it had come from, identifying that a specific employee had planted it a few days earlier."

"We referred the matter to the FBI, and we're grateful to the agents and prosecutors who moved quickly to make an arrest," the company added. "The FBI's investigation is ongoing and we will continue to support their work."

Recorder Found Behind Conference Room TV

According to the criminal complaint, OpenAI staff found the recorder on or around Sept. 30, plugged into a USB port on the back of a television and wrapped in toilet paper. The device's contents were encrypted, and the company determined it had been installed around Sept. 28.

Security cameras allegedly captured Alfozan near the conference room on several dates before the discovery, even though he had no authorization to enter the room or attend its meetings. Prosecutors say those meetings regularly involved confidential technical and strategic information about the company.

Before the device went in, Alfozan allegedly used an internal productivity tool to ask about OpenAI's building addresses, meeting rooms, room capacities, calendars, and where senior leadership and outside partners met.

Prosecutors also point to a chatbot account tied to Alfozan's spouse, which shared multiple identifiers with his own account. Between Sept. 24 and 30, that account allegedly asked how to build a continuous audio recorder, add Wi-Fi connectivity, improve live listening, and encrypt the device's SD card. The complaint does not say whose chatbot it was.

No Signs of Exfiltration, OpenAI Says

"At this time, we have no indications of this individual inappropriately accessing or exfiltrating model weights, source code, or customer data," the company said, referring to the trained parameters that make up an AI model, its underlying code, and user information. "The employee was terminated."

"We take protecting our technology, data, and people seriously, and as a matter of national importance," OpenAI said, adding that its security program is built to defend against threats ranging from insiders to cybercriminals to nation states. "This case is an example of how AI can help address these sophisticated security threats."

The case illustrates the insider half of that threat model: much of what an AI lab protects—model weights, source code, confidential technical and strategic discussions—sits behind the same doors its own employees walk through.

The company did not explain in its statement exactly what its models analyzed to identify the engineer.

Federal officials are treating the case as a national security matter. FBI Director Kash Patel said Alfozan allegedly tried to capture sensitive conversations about the company's "super intelligence technology," while U.S. Attorney Craig Missakian described the prosecution as a national security case.

Arrest Follows a String of Security Incidents

The arrest came days after OpenAI's newsroom account publicly defended the dismissal of three safety researchers—Jasmine Wang, Tomek Korbak, and Mikita Balesni—saying an investigation found they had violated clear policies on handling sensitive information.

A note from our research leaders: Last week we parted ways with Jasmine, Mikita, and Tomek after a thorough investigation found they violated clear policies on handling sensitive information. Our internal investigation uncovered a significant breach of trust beyond what's… — OpenAI Newsroom (@AINewsroom) October 9, 2026

OpenAI's security record has drawn scrutiny since July, when the company admitted that its own models escaped a test environment and breached Hugging Face's infrastructure to grab benchmark answers.

If convicted, Alfozan faces up to five years in prison, a $250,000 fine, and three years of supervised release. The charges are allegations set out in a criminal complaint; Alfozan, like any defendant, is presumed innocent unless proven guilty in court. A detention hearing is set for Oct. 19 before U.S. Magistrate Judge Alex G. Tse.