OpenAI is now training its people to treat unexpected attention with suspicion. According to a Wall Street Journal report, the company has been warning engineers that so-called “honey pot” approaches — where someone uses romance or flirtation to get close and extract secrets — are a real risk in the high-stakes AI race.
The warnings picked up fresh urgency after the FBI arrested a Saudi national and former OpenAI engineer, Abdulrahman Alfozan, for allegedly planting a hidden recording device (concealed in toilet paper) behind a television in an executive conference room.
The Journal described internal training that includes a scenario in which a stranger approached an OpenAI employee at a San Francisco playground while the employee was with their child. The stranger also had a child, made small talk, and gradually steered the conversation toward the employee’s work. The stranger was a spy.
In the same sessions, OpenAI reportedly illustrated the honey pot risk with a story about a U.S. diplomat in Russia. Spies sent a blonde woman to his hotel room with champagne. When he turned her away, they tried a redheaded woman and then a young man.
Beyond the romantic angles, the company is pushing basic operational hygiene: don’t wear OpenAI-branded clothing in public, use privacy screens on devices, and wipe whiteboards after meetings. Bribes, café eavesdropping, and other social engineering tactics are also on the list. Employees at Anthropic and Google have reportedly received similar briefings.
This is the reality of frontier AI right now. The technology is valuable enough that labs are starting to operate more like defense contractors — complete with insider-threat playbooks and training that treats ordinary social interactions as potential collection opportunities. The Alfozan arrest shows the threat isn’t purely hypothetical.
For anyone working at one of these companies, the practical takeaway is simple: be alert to people who seem unusually interested in your job, especially if the interest feels sudden or oddly persistent. The AI race has a counterintelligence side now, and OpenAI is making sure its people know it.