FBI Arrests OpenAI Engineer Over Office Recording Devices

The FBI arrested Abdulrahman Alfozan on Oct. 9 after investigators found recording devices planted in OpenAI’s offices, raising urgent questions about counterintelligence vulnerabilities at America’s most valuable artificial intelligence companies. OpenAI fired him that same day. The discovery points to a stark reality: frontier AI labs, where proprietary technology and sensitive strategic discussions are part of daily work, may not have adequate security protocols to stop insider threats.

OpenAI’s security team found the first device late last month during a sweep of a conference room used for highly sensitive discussions. It appeared to sell for about $14. Someone had plugged it into a television’s USB port and wrapped it in toilet paper. A second recording device turned up in a bag Alfozan had left in another room. Together, the discoveries suggest planning, not opportunistic surveillance.

The Investigation and Evidence

Alfozan, a 35-year-old Saudi national, joined OpenAI last year and worked on the company’s agent software. Before that, he held positions at Meta and served as a part-time partner at a Saudi-based venture capital firm. Surveillance video placed him near the compromised conference room around the time investigators believed someone installed the device, even though he didn’t have authorization to enter that space.

The Justice Department’s investigation uncovered damaging digital breadcrumbs. Queries about building a continuously recording device and encrypting its data appeared on a chatbot account associated with Alfozan’s spouse. OpenAI used its own models to study data copied from the device, giving investigators a sense of just how far the breach may have reached.

If convicted, Alfozan faces up to five years in prison, a $250,000 fine, and three years of supervised release.

Exposing a Systemic Risk

The incident has sharpened a concern security experts have quietly raised for months: spies inside AI laboratories pose a genuine national security threat. Jacob Coxon, a former Anthropic researcher who studies artificial intelligence risks, specifically warned about this possibility at both Anthropic and OpenAI. His warnings now carry the weight of demonstrated reality.

The crude method makes the case especially troubling. Espionage occurred, yet the device initially went undetected. A $14 recorder wrapped in toilet paper captured sensitive discussions until routine security protocols caught it. That isn’t the work of a sophisticated foreign intelligence operation. It shows that safeguards at even the most prominent AI firms may fall short.

The discovery also raises uncomfortable questions about hiring and vetting. Alfozan held two roles, working full-time at OpenAI while serving as a venture partner at a Saudi firm. Those roles created potential conflicts of interest that apparently weren’t flagged as problematic until after his arrest.

Why This Matters:

Artificial intelligence companies sit where economic value, technological innovation, and national security meet. OpenAI’s valuation exceeds $150 billion. The technology these labs develop will shape competitive advantage between nations for decades. When insider threats go undetected for weeks, the episode underscores that market-leading firms can’t guarantee the security of their operations without stronger institutional safeguards. The federal government’s role in investigating, prosecuting, and maintaining counterintelligence reflects a legitimate security function. But the incident also suggests private companies have been slow to adopt security practices proportionate to the value and sensitivity of what they protect. As AI competition intensifies globally, companies and policymakers must reckon with whether current security architectures can withstand determined adversaries, whether state-backed or acting independently.