October 14, 2026
Someone Hid a Bug in OpenAI’s Boardroom. OpenAI Used Its Own AI to Catch Him.
A small camera, wrapped in toilet paper, was plugged into a USB port somewhere behind a television in a conference room in San Francisco…
By Michael Koyfman
2 min read
A small camera, wrapped in toilet paper, was plugged into a USB port somewhere behind a television in a conference room in San Francisco. It's just that detail alone is enough to make you think of a bad spy movie. Stranger is the remainder of the story, which broke this week.
On Sept. 30, according to a complaint filed by the Justice Department, OpenAI discovered the device in an executive meeting room where "highly sensitive technical, commercial and strategic information" typically was shared by the meetings. The company dialed 911 to call the FBI. A few days later, on October 9, federal agents arrested a 35-year-old engineer Abdulrahman Alfozan on one of OpenAI's product teams who had joined the company in July 2025. OpenAI fired him. He's charged with one count of using a device to intercept oral communications, which carries up to five years in prison and a fine of up to $250,000. He has yet to be convicted and neither he nor his attorney has responded in public.
What is most interesting to me, at least, is how OpenAI arrived at its conclusion who did it. It has made its own models to learn about the device and backtrack its origins, and concluded that the device had been planted in the company's home office by a specific employee, it said on X. The marketing hype aside, it was a pretty interesting inversion: the lab that markets AI as a tool for analysing, uses it to analyse a piece of hardware within its own walls. OpenAI also informed TechCrunch that the device was highly encrypted and contained audio monitoring capabilities as well as a way to move data that would circumvent detection. Investigators aren't able to say what, if anything, it recorded, since it was encrypted.
The complaint is a series of small, blundering steps. Alfozan allegedly asked questions around meeting rooms, capacities, calendars and where leadership was located in the space using a productivity tool, and was seen close to the space at least six times in the last month on the CCTV footage. He was spotted by OpenAI when the company added a camera, which captured him going in and inside the TV. After he was arrested officials say they found another identical device in a bag he left behind in another room with his wallet and ID.
But, why would anyone do this? Nobody has said. The complaint doesn't detail what he was supposed to do with the audio. The largest frame yet came from FBI Director Kash Patel, who said the scheme was designed to "capture sensitive communications about the company's super intelligence technology. Pay attention to the word super intelligence — the federal law enforcement agency does not use the term. For U.S. Attorney Craig Missakian, it was a simple statement: "For us, it is a matter of national security.
OpenAI, on the other hand, is attempting to avert the worst case scenario for reading. It claims it has no evidence that he downloaded or stole model weights, source code or any customer information. It may be worth noting, but it is what any company would say, and the investigation is days old.
It's an important point of context here, since this wasn't in a quiet week. OpenAI has also recently fired three of its own safety researchers saying they were being let down, and it is said to have cancelled its intention to release its next model in October. Those stories seem to be independent and there's no mention of this arrest in the reporting. Together they create a company that's in an unusual bind, protecting something it thinks is hugely valuable.
This is the place to sit. The debate over security in AI for years has been one of model weights on the servers and nation-state hackers on the networks. If the allegations are true, this case is a reminder that the oldest attack in the book still exists: stick a person in the building and provide him with a USB port. The labs are racing to create systems they call civilization-altering; a badge reader and a guy checking behind the TV may be standing in their way of their most sensitive meetings.