By Roman Aglaganov
AI labs say they guard their model weights like crown jewels. This week showed that the conversations about them are a target too. OpenAI said it alerted the FBI after finding a hidden recording device in one of the executive conference rooms of its San Francisco office.
What happened
The device: according to federal prosecutors, it was plugged into a USB port on the back of a television. OpenAI told TechCrunch it had “advanced encryption, audio monitoring capabilities” and a way of sending data “designed to evade monitoring”. The company said it used its own AI models to help understand it.
The arrest: prosecutors say a camera OpenAI placed near the television caught a product-team employee, who had joined in July 2025, checking behind it. He was fired and arrested on 9 October, about a week after the device was found, and charged with intercepting oral communications. He faces up to five years in prison if convicted. He and his lawyer could not be reached for comment, and the charge has not been proven in court.
The damage: OpenAI said it has “no indications” that the person accessed or took model weights, source code or customer data.
What the FBI says
The purpose of the recordings is not yet clear. FBI director Kash Patel said the scheme aimed to “capture sensitive communications about the company’s super intelligence technology”, and called protecting American AI innovation one of the FBI’s highest counterintelligence priorities.
Why it matters
As AI labs become strategic assets, they face the same espionage risks as defence contractors and chip makers. Security cannot stop at encrypting model files: plans and decisions spoken in a meeting room can be just as valuable. Expect more labs to sweep rooms for devices, restrict who attends sensitive meetings and work more closely with governments on security, with all the tension that brings for open, research-style cultures.
