Safety2 min read

Building With AI Leaves a Written Trail Prosecutors Can Read

By , Senior AI ConsultantPublished

A hidden recorder in an OpenAI meeting room was allegedly built with chatbot help, and the prompts are now part of the case: AI makes building easier and keeps a record of who asked.

On October 9, federal agents arrested Abdulrahman Alfozan, an OpenAI engineer, over an audio recorder hidden in a USB port on the back of a television in the executive meeting room at the company's San Francisco office. OpenAI fired him, and prosecutors allege that a chatbot account tied to his wife was used to work out how to build the device. The complaint does not name the chatbot.

Each question looks ordinary on its own

According to the complaint, the account asked how to build a continuous audio recorder, how to set up its Wi-Fi, how to improve audio and live listening, and how to encrypt the SD card. Almost everyone who asks those questions is a podcaster, a researcher or a parent with a baby monitor. Nothing in one prompt says "spy", so nobody can expect a chatbot to know who is asking.

What gives the plan away is the sequence. The same account also asked about the device's custom header and its unique network address, on the day before the recorder was installed in late September and on the day itself. A prompt that matches a physical object found in the room is what turns a chat log from background into evidence.

For a small business this is the useful half. A person who writes software but has never built hardware used to need a friend who had. Now a stockroom manager can describe a sensor that counts boxes and get a working plan. The recorder in this case was allegedly built the same way.

The record does what an informant used to do

The complaint says the wife's account shared several identifiers with his own account, so borrowing another person's login did not separate him from his prompts. The provider keeps the history with dates and times, and a court order can turn it over.

The Palisades fire case in Los Angeles shows the same pattern. In October 2025 prosecutors charged a man with starting that fire, and the complaint cites his ChatGPT use: a request for a picture of a burning forest, and a question, asked during his 911 calls, about whether you are at fault if a fire is lit by your cigarettes. Both cases are allegations, not verdicts, but they point the same way. Chat history is becoming what search history became after the Denver case: something investigators ask for early.

I expect the same requests in ordinary disputes, such as a fired employee's lawsuit, a divorce or a fraud claim. The record cuts both ways, because a history that shows an ordinary reason for a question clears the person who asked it. Most users are in that group, and for them the lesson is only to know the record exists.

What a business can do about it

OpenAI's staff found the recorder about two days after it was installed. A recorder like this is made from cheap parts, so any meeting-room screen with a spare USB port is a place to hide one, and covering or disabling those ports costs almost nothing.

The second step costs a sentence. Tell staff plainly that what they type into an AI tool at work is a record, kept with their name on it, that the company and a court may read later.

Share this

STAY INFORMED

Get AI intelligence like this delivered to your inbox.

Free forever · Unsubscribe anytime


You May Also Find Valuable