Someone allegedly tried to spy on OpenAI using AI. That’s the short version. The longer version involves toilet paper, a USB port behind a TV, and chatbot prompts asking how to encrypt an SD card.
According to Engadget, US federal prosecutors have arrested and charged Abdulrahman Alfozan, a 35-year-old Saudi national who had been working as an engineer at OpenAI since July 2025. The charge: allegedly hiding a recording device inside an executive conference room at OpenAI’s San Francisco headquarters. The room, which Alfozan was not authorized to enter, regularly hosted meetings about the company’s product roadmap, ongoing research, and government engagement. He reportedly concealed the device, a DIY build, using toilet paper and plugged it into a USB port on the back of a wall-mounted TV. OpenAI found it on September 30.
The details around how Alfozan allegedly built and installed the device make this case genuinely unusual. Prosecutors say he may have used a chatbot, on an account linked to his spouse, to ask how to “build a continuous audio recorder” and “encrypt the SD card.” He also reportedly used a productivity tool to research OpenAI’s internal facilities layout. So the alleged plan to spy on an AI company was, at least in part, assisted by an AI tool. The irony is hard to miss.
OpenAI’s response is also notable. The company said it used its own AI models to help analyze and understand the device. In a statement posted to X, OpenAI wrote that it has “no indications of this individual inappropriately accessing or exfiltrating model weights, source code, or customer data” and confirmed the employee was terminated. That’s a carefully worded statement, and it suggests OpenAI’s internal security team is now treating AI as part of its investigative process, not just its product line.
The case matters beyond the odd specifics. OpenAI is at the center of some of the most commercially and geopolitically sensitive AI development happening right now. Its work with government agencies, its model roadmap, and its research directions are exactly the kind of intelligence that state-linked actors or well-funded competitors would want access to. Whether or not Alfozan was acting on behalf of a foreign government has not been established publicly, but the profile of the alleged target, an executive conference room used for roadmap and government discussions, points to someone who understood what they were looking for.
On October 5, OpenAI security found a bag elsewhere in the office containing Alfozan’s wallet, his ID, and a second recording device. He made his first court appearance on Tuesday. If convicted, he faces up to five years in federal prison, a $250,000 fine, and three years of supervised release.
For developers and founders building on or around AI infrastructure, this is a reminder that physical security is part of the threat model now, not just data access controls and API key hygiene. The most sensitive information at a company like OpenAI often lives in conversation, not code repositories.
