A listening device worth about $13, hidden behind a TV in an OpenAI executive conference room, has become a warning for the entire AI industry. The ChatGPT maker’s case against a former engineer, now charged by US federal prosecutors, shows how valuable AI secrets have become and how cheaply someone can try to reach them.OpenAI is also not alone in bracing for this. According to a report by The Wall Street Journal, employees at Google and Anthropic have received similar warnings about spies, honey traps and strangers who ask too many questions. Security experts quoted in the report believe this arrest is unlikely to be the last.
OpenAI says its own AI models helped catch the alleged office spy
Abdulrahman Alfozan, a 35-year-old Saudi national and US permanent resident, joined OpenAI as an engineer in July 2025. Prosecutors allege he wrapped a Seeed Studio XIAO ESP32-S3 Sense, a thumb-sized board with a microphone and camera, in toilet paper and plugged it into a USB port behind the TV.Only executives and their assistants could book the room. It hosted leadership meetings, product sessions and briefings with government officials. OpenAI found the device during a security sweep on September 30 and says it used its own models to trace it to the employee. The FBI arrested him on October 9.He faces one charge of intercepting oral communications, which carries up to five years in prison. OpenAI says it has no sign of him accessing model weights, source code or customer data. His lawyer called the allegations “wholly out of character”, adding that his client is not charged with stealing trade secrets.
OpenAI warns employees about playground spies and honey traps
As per WSJ, OpenAI’s internal training has been preparing staff for moments like this. One scenario describes a stranger at a San Francisco playground who befriends an employee there with their child, makes small talk and slowly steers it toward work. The stranger is a spy.Employees are told not to wear company-branded clothing outside office, to use privacy screen protectors on public transport and to wipe whiteboards after meetings in case cleaning staff are compromised. Some sessions cover “honey pots”, where agents use romance to get close to a target. Staff at Google and Anthropic have been briefed along similar lines.
Trillion-dollar valuations make Google, Anthropic and OpenAI bigger spy targets
The money explains the nerves. OpenAI and Anthropic are each expected to seek valuations of over $1 trillion when they go public. Phil Venables, former security chief at Google Cloud, told WSJ that the huge cost of training foundation models makes them even more valuable intellectual property.Rapid hiring adds to the risk. AI labs have added thousands of people in a short time, and engineers hop between rivals often. “Catching one guy means there are half a dozen out there,” Randy Watt, president of security consultancy SRW, told the publication. Aviv Nahum, CEO of insider-risk startup Above Security, questioned whether security checks are keeping pace with that growth.The industry has been here before. In January, a former Google engineer was convicted of stealing AI trade secrets for China. In 2022, a former Twitter employee was jailed for spying for Saudi Arabia.OpenAI caught its device within days of it being planted. For Google, Anthropic and every other lab chasing a trillion-dollar valuation, the bigger worry is what their own sweeps have not found yet.
