An OpenAI engineer has been accused of hiding a recording device in an executive conference room at the company's San Francisco offices and now faces federal charges.

Adding a surprising turn to the case, prosecutors say AI technology assisted the suspect in his preparations, while OpenAI says its own models helped investigate the matter.

What prosecutors say Alfozan did

According to the Department of Justice, Abdulrahman Alfozan, 35, hid the recording device inside toilet paper and attached it to the USB port of a television set on September 28, 2026. OpenAI's security team discovered the device during a search on September 30.

WIRED identified the gadget as a $13.99 Seeed XIAO ESP32-S3 Sense board — a commercially available development board — equipped with encrypted microSD storage. After reviewing surveillance clips, authorities discovered a second recorder on October 5 inside an unattended bag containing Alfozan's identification.

Alfozan was arrested on October 9 and brought to court on October 13. If found guilty, he faces five years in prison and a $250,000 fine. According to the New York Times, another hearing is scheduled for October 19. Prosecutors have not provided any information regarding a possible motive.

The chatbot trail, and OpenAI's AI-assisted response

Prosecutors claim that Alfozan used a workplace productivity agent to search executive meeting rooms and schedules. They further allege that a chatbot account linked to his spouse was used to check for around-the-clock recording capability, Wi-Fi access, live listening, and SD card encryption.

OpenAI said in a statement on X that its models allowed security staff to understand the device and identify who planted it. AI was only part of the investigation, however: camera footage, chat records, and physical evidence also assisted investigators.

The company said it has found no evidence to suggest that model weights, model source code, or customer data were accessed or stolen without permission.

What a single insider can cost

Had confidential data been compromised, the situation would have become significantly worse and far more expensive. IBM states that data breaches cost businesses an average of $4.99 million globally, and organizations that use security AI and automation tools have reported breach costs $1.93 million lower than companies that do not.

Insider attacks create additional financial risks. The Ponemon Institute and DTEX estimate that the average cost of containing an insider breach incident is $247,587, and for companies that took more than days to contain an incident, the average cost is $21.9 million. These figures reflect average industry costs, not losses attributed to OpenAI.

Why budgets and AI oversight are climbing

PwC found that 84% of surveyed security and finance leaders expect cyber budgets to rise, while 58% of security leaders prioritize AI spending.

Gartner projects that AI governance-platform spending will rise from $492 million in 2026 to more than $1 billion by 2030. BCG found that nearly half of companies already see value from AI, yet only 5% have complete controls in place for autonomous agents — a gap that is relevant to the kind of workplace AI agent cited in the allegations.

A separate fight over watching the models

Separately, Cryptopolitan reported that three former OpenAI researchers have questioned how the company monitors advanced models and have called for independent safety reviews. That debate concerns model behavior rather than the alleged bugging, but both stories show how quickly trust and oversight are becoming business issues for AI companies.