FBI Arrests OpenAI Engineer Over a Hidden Conference-Room Recorder: the Company Says Its Own AI Models Helped Identify Him

Federal prosecutors have charged a 35-year-old OpenAI engineer with planting a covert audio recorder inside an executive conference room at the company's San Francisco headquarters — a case that reads like corporate espionage, and one that OpenAI says was cracked in part by its own AI models. Abdulrahman Alfozan was arrested on Friday and made his initial appearance in federal court on Tuesday, according to the Justice Department.

The device itself was small and deliberately inconspicuous. Court documents describe a thumb-sized unit with a microphone that could also connect to a camera, plugged into the USB port on the back of a television in a conference room where meetings routinely covered highly confidential, nonpublic information. It was concealed inside toilet paper, installed on or about Sept. 28, and discovered two days later during a security sweep. Its stored data was encrypted and inaccessible to investigators.

OpenAI, in a statement on X, said it used its internal models to understand the device and where it had come from, identifying the specific employee who had planted it a few days earlier. The company said there are no indications the individual improperly accessed or exfiltrated model weights, source code, or customer data, and confirmed the employee was terminated on Friday. It has referred the matter to the FBI, which is continuing to investigate.

The complaint paints a picture of methodical preparation. Alfozan joined OpenAI as an engineer in July 2025. Security footage allegedly placed him near the conference room six times last month despite having no authority to access the space; after the company added a camera to the room, it captured him entering and checking behind the television where the device sat. Prosecutors also say he used an AI productivity agent for about a month to ask about building addresses, meeting rooms, room capacities, calendars, and the locations of executives.

The filing goes further: his spouse allegedly submitted prompts to a chatbot about building, customizing, installing, audio recording, testing, obfuscating, and remotely accessing a device — including requests like building a continuous audio recorder, customizing Wi-Fi connectivity, improving live listening, and encrypting an SD card. The same account made inquiries referencing unique identifiers of the device found at OpenAI. A second, identical recorder turned up in a bag Alfozan left unattended in a different room on Oct. 5, along with his wallet and identification.

He is charged with using or endeavoring to use an electronic, mechanical, or other device to intercept oral communications, an offense carrying up to five years in prison, a $250,000 fine, and three years of supervised release. FBI Director Kash Patel said Alfozan "allegedly concealed a covert recording device inside his employer's executive conference room to secretly capture sensitive communications about the company's super intelligence technology." A LinkedIn profile indicating studies at MIT and UC Berkeley and prior stints at Meta and Microsoft could not be verified by Reuters; Alfozan has been a US legal permanent resident since September 2020, and the Saudi embassy did not respond to a request for comment.

The arrest lands during an unusually turbulent stretch for OpenAI's security story. On Oct. 9 the company fired three researchers after an investigation into the handling of sensitive information — a dismissal the three publicly disputed. Independent reporting on agents reaching government and public-interest websites has also kept the company's containment practices under scrutiny, even as Reuters reported that OpenAI told investors its annualized revenue for September was almost $50 billion.

What it means: frontier labs now hold assets — model weights, safety roadmaps, product plans — that nation-state-grade actors would plausibly want, and the defense perimeter has to extend to a TV's USB port. The most consequential detail may be OpenAI's claim that its own models helped identify both the device and the person who planted it. If that holds up, it is a small but pointed example of AI being used to counter the security threats that come with building it.