Around September 30, OpenAI’s security team found a thumb-sized audio recorder wrapped in toilet paper and plugged into a USB port behind a television in an executive conference room in one of its San Francisco buildings — a room, the complaint says, reserved for leadership meetings and briefings with government officials. Within days the FBI arrested a 35-year-old OpenAI engineer, Abdulrahman Alfozan, and charged him with trying to secretly record conversations using a device. He is presumed innocent and, through counsel, says he will fight the allegations.
I write a blog about the government collecting your data, so why am I spending a Thursday on a private employee and a bug behind a TV? Because the charge lands squarely on the statute this entire project is built on. Alfozan is charged under the Wiretap Act — Title III, 18 U.S.C. § 2510 and following — with using, or endeavoring to use, a device to intercept oral communications. The same law I keep holding up as the blueprint for restraining the government is, at its foundation, a criminal prohibition that binds everyone. That is worth sitting with for a minute.
Start with what the model act calls prohibition-first. My § 2 makes interception of protected content illegal by default, allowed only through a narrow, justified authorization. People tend to read that as a leash on government. Title III shows the instinct is bigger than that. Its baseline is not ‘the government needs a warrant’; it is that capturing someone’s private communications without authorization is a crime, full stop, whoever does it. Alligator clips on a phone line, a recorder behind a television, or — the thing I spend most of my time worrying about — a government query run against a database: same genus of wrong, different actor and different scale. The prohibition is the foundation; authorization is the exception carved out of it. This case is the foundation doing exactly the job it was built for.
Notice, too, what the statute protects. The device was aimed at a conference room, but the protected thing is not the room — it is the conversation, the reasonable expectation of privacy in what people say to each other, the interest Katz put at the center of Fourth Amendment law. That is the first principle of my § 1 definitions: protection attaches to content, in whatever form it is captured. A leadership meeting recorded to an encrypted memory card is the same protected content whether the instrument is a reel-to-reel in 1968 or a thumb drive in a toilet-paper roll in 2026. The container changed. The thing the law protects did not.
Then there is the device itself, and this is where the case turns modern in a way that should worry anyone drafting in this space. Title III does not only forbid the interception; it regulates the instruments — § 2512 reaches surreptitious interception devices. But look at what the instrument has become. Per the complaint, this one was thumb-sized, built for Wi-Fi, with an encrypted memory card and features meant to hide its own activity. And here is the detail that stopped me: a chatbot account linked to the defendant’s wife was allegedly used to ask how to build a continuous audio recorder, add Wi-Fi, encrypt the card, and conceal its operation, and he allegedly asked an AI tool to list OpenAI’s meeting rooms and the locations for ‘leadership and exec.’ The knowledge to design and place a covert interceptor is now a handful of prompts away. That is the same collapse of friction I keep flagging in other contexts: the practical limits that used to decide who could build and deploy surveillance tools are evaporating, and the statute’s hardware-era assumptions evaporate with them.
There is an irony here that ties straight back to last week. Last Thursday I wrote about the government conscripting seized data into AI training models — surveillance pointed through AI. This week the vector flips: the AI lab is the target, and AI tools were allegedly used to build the bug aimed at it. The watchers’ tools are being turned on the people building the next generation of watchers. And watch the government’s framing of it — the Director of National Intelligence and the FBI Director cast an alleged bug in a private company as a national-security matter, a question of protecting American AI innovation. The old lines between private corporate secrets, government interest, and foreign intelligence are dissolving in real time.
Which is exactly why the next paragraph needs precision, because the headline runs well ahead of the charge. Alfozan is a Saudi national; the press reports that he advised Saudi organizations, that a Riyadh venture firm lists him as a partner alongside a member of the royal family, and that he and relatives had booked one-way tickets to Saudi Arabia. Those are associations and reporting. The government has not charged him with espionage, and the Justice Department’s own release does not allege that the Saudi government directed any of this. He is charged with is a single Wiretap Act offense — intercepting, or endeavoring to intercept, oral communications with a device — carrying up to five years. OpenAI itself said he did not appear to have extracted sensitive information. A complaint is an accusation, not a verdict. The distance between ‘charged with a wiretap crime’ and the ‘foreign-spy’ story the facts invite is precisely the gap this blog exists to keep straight — alleged is not adjudicated, and nationality is not a charge.
So what does a sixty-year-old wiretap statute, used to arrest a man for a recorder behind a TV, tell us about the digital-property law I am drafting? First, that building on a prohibition-first spine is not naive — it works. The reason the FBI could walk into that building and make an arrest is that Title III drew a clear default: capturing protected content without authorization is wrong, and it is wrong whether a citizen, an employee, or an agency does it. That is the proof of concept for the whole model act — a baseline clear enough and broad enough to be enforced against anyone.
Second, that the baseline must be written for the world the bug came from, not the world Title III was drafted in. A statute built around telephone taps and hand-placed room bugs now is required to reach thumb-sized encrypted AI-assisted recorders, insider threats inside companies holding secrets that governments covet, and a national-security frame that treats a startup’s conference room like a secure government facility. The lesson for drafting is the one I keep returning to: define the protected content by function, broadly enough that it does not matter whether the interceptor is a wiretap, a thumb drive, or a database query; and define the ‘device’ by what it does, so the law is not forever chasing hardware it cannot keep up with.
The bug in the toilet-paper roll is a small and slightly absurd story. But it is a clean window onto the first principle of this entire project: before you ever get to warrants, minimization, retention, and remedies, there has to be a baseline that says capturing someone’s communications without authorization is simply not allowed. Title III has held that line — against the government and against the private actor alike — for almost sixty years. The work now is to hold it for the database, the model, and whatever the next thumb-sized thing turns out to be.
Leave a Reply