Someone bugged OpenAI's boardroom with a device wrapped in toilet paper

Somewhere in OpenAI's San Francisco office there is a television mounted in a conference room that executives use for the conversations they would least like to read about in the press. Research plans. Product plans. Meetings with partners and, according to the FBI, meetings with government officials. On September 30, the company's technical surveillance countermeasures team swept that room and found something plugged into a USB port on the back of the TV.

It was an audio recorder. Someone had wrapped it in a small piece of toilet paper so it would blend into the cable mess behind the screen. OpenAI told Fortune the device had encryption, audio monitoring and a data transfer mechanism built to slip past the company's network monitoring. It had been sitting there for about two days.

On October 9, the FBI arrested Abdulrahman Alfozan, a 35-year-old OpenAI engineer, and on Tuesday he made his first appearance in federal court in the Northern District of California. He is charged with using or attempting to use a device to intercept oral communications. He has not been convicted of anything, and everything below comes from the criminal complaint, the Justice Department and OpenAI. None of it has been tested in court.

What the complaint says happened

The timeline in the complaint reads like a heist written by someone who has never seen a heist movie.

Alfozan joined OpenAI in July 2025 and worked on what Fortune describes as the company's "Agents Platform." According to court documents reported by CBS News and Fortune, a chatbot account linked to his spouse started receiving prompts on September 24 about how to build, install, test and remotely access a continuous audio recording device. The questions reportedly got specific: custom MAC addresses, iPhone hotspot profiles, live listening sessions, and how to build a second and third device.

Prosecutors also say he used an internal productivity tool to look up OpenAI building addresses, meeting rooms and their capacities, calendars, and where "leadership and exec" meetings and external partner meetings would take place. In other words, the alleged spy used the company's own software to plan the bugging of the company.

The device went in on or about September 28, according to the Justice Department. Security found it two days later and called the FBI. Then, the complaint says, OpenAI waited.

On October 5, a camera allegedly caught Alfozan walking into the conference room and looking behind the television. He did not have authorization to be in that room, prosecutors say, and footage showed him near it several times before the device turned up. The same day he allegedly left a bag unattended in a different room. OpenAI security opened it and found his ID, his wallet and a second recording device similar to the one behind the TV. Fortune reports both devices were from the same maker, Seeed Studio, which sells small development boards to hobbyists.

The next day, according to Fortune, he reported the bag's contents missing: a credit card, about $500 in cash, AirPods, headphones, sunglasses and a power bank. The recorder was not on the list.

Fortune also reports that he had been booked on a one-way flight to Saudi Arabia, via Istanbul, with his spouse and another family member. The FBI picked him up on October 9. OpenAI fired him the same day, timing the termination with the bureau so it would not tip him off.

The AI that caught the insider

OpenAI's account of its own response is the part of this story that will get quoted at conferences for the next two years. The company says it used its internal AI models to identify the person responsible and trace where the device came from. It has not said how. We don't know whether that means a model chewing through badge logs and camera footage, or someone pasting a photo of a circuit board into ChatGPT and asking what it was.

Whatever the method, it lets OpenAI tell a tidy story: our models are so good they even catch the people trying to steal them. That is a convenient story for a company that sells those models. It is also the company's claim, and it hasn't been independently checked.

OpenAI says it took "gradual steps to contain the risk while the investigation continued," per NBC News. It restricted his access, cut him off the network, put him on leave and took his equipment. The company also says it has evidence he "intended to deploy other recording devices." And the line every OpenAI investor wanted to read: "At this time, we have no indications of this individual inappropriately accessing or exfiltrating model weights," the company said, adding that it also had no sign of source code or customer data leaving the building.

Note the shape of that sentence. "No indications" is what you say when you have looked and found nothing. It is not the same as "nothing happened." The FBI investigation is still open.

Washington is very interested

The government is treating this as a national security case, and it is saying so loudly. FBI Director Kash Patel posted that 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." He also wrote that "Protecting American SI innovation is critically important," which may be the first time the head of the FBI has used "SI" as if everyone knows what it means.

Attorney General Todd Blanche weighed in on X too, according to Fox Business. Nobody has publicly said who, if anyone, Alfozan was working for. The charge itself is an eavesdropping charge, carrying a maximum of five years in prison, a $250,000 fine and three years of supervised release. There is no espionage count and no allegation in the public record that he passed anything to a foreign government.

His lawyer, William Pernik, told CBS News his client is presumed innocent and warned against "jumping to conclusions at this very early stage of the case." He also asked people not to "politicize" the prosecution based on his client's nationality. That warning will be ignored by a lot of people online, and it shouldn't be. As of today the evidence the public has seen is a bugged room, a bag, some chatbot prompts and a flight booking. Motive is still a blank.

Why a bug in a TV matters more than it looks

The AI industry has spent three years worrying about the wrong kind of leak. The fear has been model weights walking out the door on a thumb drive, or a nation-state hacking team getting into the training cluster. Labs built sophisticated defenses for that. OpenAI's own security story has leaned heavily on protecting weights.

This case, if the allegations hold, went around all of it. The target was a conversation. The tool was a hobbyist board you can order online. The cover was toilet paper. The planning allegedly happened on a consumer chatbot and the employer's own scheduling software. The most valuable thing in a frontier lab may be what the people at the top of it say to each other, and to government officials, in a room with a TV.

OpenAI did catch it, and fast. It runs a dedicated counter-surveillance team that sweeps executive rooms, which tells you the company already thought this was a realistic threat. Most companies of its size don't have one. Expect a few of its competitors to start hiring for one.

Alfozan's detention hearing is set for October 19. Until then, every AI lab in the Bay Area has a new line item on the facilities budget: someone to look behind the TV.

Sources: Fortune, NBC News, CBS News, Fox Business.