Spies and A.I.

“Corporate espionage comes for OpenAI

 

Jacob Coxon, a former Anthropic researcher who warned about the risks of artificial intelligence, has raised alarms about the possibility of spies inside Anthropic and OpenAI.

 

OpenAI’s revelation yesterday that it had discovered a hidden recording device in one of its offices and traced it to an employee, underscores the potential reality of that threat — and adds to concerns around frontier A.I. labs, Sri Muppidi reports.

 

What happened:

 

    An OpenAI security team discovered the recording device late last month while sweeping a conference room used for highly sensitive discussions, according to the Justice Department. (The device, a tool that appears to sell for about $14, was plugged into the USB port of a TV and wrapped in toilet paper, the department said.)

 

    Investigators traced the device to Abdulrahman Alfozan, an engineer who joined OpenAI last year, the Justice Department said.

 

    Surveillance video showed Alfozan, who didn’t have authority to use the room, near the space around the time that the device was suspected of being installed.

 

    They also found another recording device in a bag that he had left in another room, according to the Justice Department.

 

    Investigators also discovered queries about how to build a continuously recording device and encrypt its data made on a chatbot account associated with his spouse. OpenAI said the company had used its models to help study data copied from the device.

 

    The F.B.I. arrested Alfozan on Oct. 9, and OpenAI said it had fired him that day.

 

Who is Alfozan? The engineer, 35, is a Saudi national, according to the Justice Department. He worked on OpenAI’s agent software and previously worked at Meta, according to a LinkedIn profile that appears to be his. He was also a part-time partner at a Saudi-based venture capital firm.

 

He could face up to five years in prison and a $250,000 fine, then three years of supervised release if he is convicted.

 

Others have been charged with spying on tech companies before, including former employees of Twitter and Apple.

 

Washington and Silicon Valley already view the U.S. lead in A.I. as a top national security issue, and competition among labs is fierce.

 

Corporate espionage is only the latest safety and security issue A.I. labs are confronting. OpenAI has acknowledged several instances of its software going rogue, with agents hacking companies and meddling with government websites. (It was also criticized after it fired three safety researchers for sharing sensitive information with what they said was a third-party A.I. safety group.)

 

This week, Anthropic, which Coxon criticized about its commitment to safety, warned of a different threat: that its models could be months away from self-replicating and possibly hacking smart home devices.

 

    We’ve also written before about the potentially huge liability risks Anthropic and OpenAI could run into if their products cause significant harm.

 

Safety is another consideration for potential investors in the companies’ planned I.P.O.s. The topic may come up today when Anthropic hosts an investor day for its planned stock sale.

 

    Anthropic executives may also be asked about the company’s latest financials — including how they calculate revenue.” [1]

 

1. DealBook: Spies and A.I. New York Times (Online) New York Times Company. Oct 14, 2026.