OpenAI says its own models helped trace a hidden office bug

Federal prosecutors have charged OpenAI engineer Abdulrahman Alfozan, 35, over a recording device allegedly hidden in an executive conference room at the company’s San Francisco offices. OpenAI said on Twitter on October 13 that it used its own models to understand the device and where it came from, identifying the employee who had planted it a few days earlier. The charge is using or attempting to use a device to intercept oral communications, and OpenAI says it has no indication that he improperly accessed or removed model weights, source code or customer data.
The recorder turned up during a sweep by OpenAI’s technical surveillance countermeasures team, the staff who check rooms for listening devices, the company says; the complaint dates the discovery to September 30. It was plugged into a USB port behind a television and concealed in toilet paper, with its contents encrypted, and OpenAI determined it had been installed around September 28, prosecutors say. They describe the room as one frequently used for discussions of confidential research, product plans, business strategy, government engagements and commercial relationships. OpenAI calls the device sophisticated and designed to evade detection, with advanced encryption, audio monitoring and a way of transferring data meant to avoid monitoring.
Prompts, footage and a second device
Surveillance footage showed Alfozan near the conference room on multiple occasions before the discovery, the complaint alleges, though he had no authorization to enter it or attend meetings there. Before the device was installed, prosecutors say, he used a productivity tool to ask about OpenAI’s building addresses, meeting rooms, room capacities and calendars, and about the locations of leadership and outside partner meetings.
Prosecutors also cite a chatbot account tied to his spouse that shared identifiers with his own account. Between September 24 and 30, the complaint says, it submitted prompts about building, installing, obscuring and remotely accessing a recording device, among them how to “build a continuous audio recorder,” how to add Wi-Fi and listen remotely, and how to “encrypt the SD card.” On September 27 and 28, by prosecutors’ account, the same account asked about the specific device later recovered, including its unique MAC address.
On October 5, the complaint alleges, Alfozan entered the conference room and appeared to look behind the television. That same day, prosecutors say, security personnel opened a bag he had left unattended in a different room and found his wallet, his ID and a second recording device similar to the first. OpenAI, for its part, says it has evidence that he intended to place other recording devices in sensitive spaces to secretly record meetings.
What OpenAI did after September 30
OpenAI presents the case as an example of how AI can counter sophisticated security threats, though what it credits its models with is narrower and forensic: employees used them to work out what the device was and where it came from, and investigators used them to analyze forensic images of the device’s storage and assess changes in the underlying data.
Between September 30 and the arrest, the company says, it gradually restricted Alfozan’s access to systems and spaces, then suspended his physical and network access, placed him on leave and secured his company-issued equipment. It coordinated the timing of his firing with the FBI to avoid disrupting the investigation and fired him on October 9, the day he was arrested. OpenAI says he was a member of its product team; a LinkedIn profile matching his name lists him as engineering lead on the company’s AI agents platform, and prosecutors say he had worked at OpenAI since July 2025. In its statement, OpenAI thanked the agents and prosecutors who “moved quickly to make an arrest” and said the FBI’s investigation is ongoing.
OpenAI has taken an employee matter to the FBI before: The Wall Street Journal reported in January 2024 that the company reached out to the bureau after a forensic investigation of a former employee’s laptop raised suspicions that the employee had taken company secrets to China. That employee was later exonerated, according to a person familiar with the matter.
Up to five years
Alfozan, a Saudi national, made his initial appearance in federal court on October 13. If convicted, he faces up to five years in prison, a $250,000 fine and three years of supervised release. The National Security, Cyber & Special Prosecutions Section of the U.S. Attorney’s Office for the Northern District of California is prosecuting the case, and U.S. Attorney Craig Missakian said in a statement: “For us it is a matter of national security.” FBI Director Kash Patel said protecting U.S. AI innovation is among the bureau’s highest counterintelligence priorities.
He remains in federal custody, and a detention hearing before U.S. Magistrate Judge Alex G. Tse is set for October 19.
Up next
Gemini 4 Argon lands fifth on Arena, behind Opus 5.5
“5th spot” is where Gemini 4 Argon landed on Arena’s overall ranking, per a Threads post, trailing all three rivals Google picked for its own benchmark table. Has a vendor’s benchmark table ever matched how a model felt in your own work?
Related stories
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

