The company that says it is building super intelligence found a recorder hidden in toilet paper on the USB port of the TV in its executives'' room. Whoever planted it, per the federal charge, is one of OpenAI''s own engineers. And he asked a chatbot how to build the bug.
I will go by what the Department of Justice published, because the rest is imagination.
The case timeline
The device was installed around September 28 in OpenAI''s executives'' room in San Francisco. Employees found it on September 30: plugged into the back of the TV, covered by toilet paper, encrypted content. The arrest came on October 9.
The accused is Abdulrahman Alfozan, 35, an engineer at OpenAI since July 2025. The charge is using a device to intercept oral communications. If convicted: five years and a 250 thousand dollar fine.
The room was not just any room. It frequently handled proprietary research, product plans and non-public information about technical progress, priorities and government engagements.
The first failure: access control
He had no authority to enter that room, and cameras showed him near it on several dates. Before planting the device, he searched an internal tool for rooms, calendars and executives'' locations.
Physical access without screening, corporate schedules exposed to people who do not need them, and executives'' locations queryable internally: each of those is a control that exists on paper and that the case showed broken in practice.
The second failure: the conversation with the model is recorded
A chatbot account associated with his wife, which shared several identifiers with his own account, sent prompts between September 24 and 30 about building, installing, testing, obfuscating and remotely accessing a device: a continuous audio recorder, customizable Wi-Fi, live listening and SD card encryption.
The same chatbot asked about the custom header and the unique MAC address of the device found in the room. Let that sink in: the man allegedly asked an AI model how to build the bug, and then asked the model for the bug''s MAC address. His questions are now federal evidence.
The part that closes the case
It came on October 5. Cameras showed him returning to the room to look behind the TV, after the discovery. That same day, he left a bag unattended in another room: inside, along with his wallet and ID, a second identical recorder.
OpenAI''s response: it says it used internal AI models to identify the person responsible and trace the origin. The alleged bug designed with AI help was dismantled with AI help. There is no indication of unauthorized access to model weights, code or customers.
The government did not treat this as a police report. The director of national intelligence spoke about winning the super intelligence race, and the FBI''s director called the case a counterintelligence priority.
Conclusion
Two lessons for anyone operating technology. The physical layer is still the one nobody models: the industry''s most watched perimeter was beaten with toilet paper and a USB port. And the conversation with the model is recorded: when the question is how to build a bug, it stops being a draft and becomes testimony.