A $13 bug, toilet paper, and a Saudi engineer: OpenAI's spy case rattles all of Silicon Valley

The device cost about $13. It was wrapped in toilet paper and plugged into a USB port behind a television. The room it was hidden in hosted OpenAI leadership meetings, product sessions, and briefings with government officials — and only executives and their assistants could book it. That gap, between the value of what was being discussed and the cheapness of the tool used to listen in, is now the central anxiety of the AI industry.

The FBI took Abdulrahman Alfozan into custody on October 9. Alfozan, a 35-year-old Saudi national and US permanent resident, had come on board at OpenAI as an engineer in July 2025. Federal prosecutors allege he concealed a Seeed Studio XIAO ESP32-S3 Sense — a thumb-sized board equipped with a microphone and camera — behind a TV in an executive conference room at OpenAI's San Francisco headquarters. During a security sweep on September 30, OpenAI found the device and says it deployed its own AI models to link it to Alfozan.

A single count of intercepting oral communications — an offense punishable by up to five years in prison and a $250,000 fine — has been brought against Alfozan. He is not charged with stealing trade secrets or acting as a foreign agent. OpenAI says it has found no evidence he accessed model weights, source code, or customer data.

Protecting confidential information about AI development was essential to safeguarding our nation's economic strength and national security, and the FBI views that mission as one of our highest counterintelligence priorities.— Kash Patel, FBI Director

Alfozan's lawyer described the allegations as "wholly out of character" and pointed out that his client faces no charges related to trade-secret theft, according to the Times of India.

The arrest has given fresh urgency to internal security training that OpenAI had already been running. The Wall Street Journal, as reported by multiple outlets, described one training scenario in which an OpenAI employee at a San Francisco playground is approached by an unknown person who chats casually before gradually turning the conversation toward the employee's work. The stranger is a spy. Another scenario, drawn from a real diplomatic incident, describes a US diplomat in Russia who was approached sequentially by a blond woman, then a redheaded woman, then a young man — each sent to his hotel room by a foreign intelligence service. These are the so-called 'honey pot' schemes: the use of romance or seduction by foreign operatives to extract secrets from targets.

Staff are also instructed to avoid wearing OpenAI-branded clothing in public, to fit privacy screen protectors on their phones and laptops when using public transport, and to clear whiteboards after meetings on the chance that cleaning staff could be compromised. Security experts have also flagged the risk of spies eavesdropping on employees in public spaces like cafes, or offering outright bribes.

OpenAI is not alone in issuing these warnings. The Wall Street Journal, as cited by the Times of India and others, reported that staff at Google and Anthropic have received comparable briefings covering spies, honey traps, and individuals who probe for sensitive information.

The money explains the nerves. According to the Times of India, both OpenAI and Anthropic are anticipated to pursue valuations exceeding $1 trillion at the time of their public offerings. Phil Venables, former security chief at Google Cloud, told the Wall Street Journal that the enormous cost of training foundation models makes them especially valuable intellectual property. Rapid hiring compounds the risk: AI labs have added thousands of employees in a short period, and engineers frequently move between rival companies.

Catching one guy means there are half a dozen out there.— Randy Watt, President, SRW security consultancy

Aviv Nahum, CEO of insider-risk startup Above Security, raised doubts about whether the industry's security vetting processes are advancing quickly enough to match its rate of growth, according to the Times of India.

The AI industry has faced similar cases before. A former Google engineer was found guilty in January of taking AI trade secrets to benefit China. According to the Times of India, a former Twitter employee was sentenced to prison in 2022 for conducting espionage on behalf of Saudi Arabia.

OpenAI caught the device within days of it being planted — and says its own AI tools helped identify the culprit. For every other lab racing toward a trillion-dollar valuation, the harder question is what their own security sweeps have not yet found.

Why it matters — The case shows that the most valuable intellectual property in the world — AI models worth trillions — can be targeted with a $13 gadget and a roll of toilet paper, and that no AI lab, however sophisticated, is immune.

Reported by aiweekly.co, ndtv.com, nypost.com, timesofindia.indiatimes.com