OpenAI Trains Staff to Spot Seductive Spies, Confirming Silicon Valley Is Now a John le Carré Novel With Standing Desks

If someone at the gym suddenly finds you fascinating and wants to hear all about your transformer weights, OpenAI would like you to think twice.

According to The Wall Street Journal, OpenAI has been running training sessions warning employees that competitors and state-backed actors may try to get company secrets out of them. The methods include offering money, eavesdropping in cafés, and seduction, the classic “honey pot.” Staff at Anthropic and Google have reportedly had similar warnings.

The playground case

One scenario shown to employees, the Journal reports, involves a San Francisco playground. A stranger who regularly turned up with a child befriended an OpenAI employee who was also there with their kid. The small talk gradually moved towards the employee’s job. The stranger was a spy. Parents everywhere now have proof that the other parent who keeps asking “so what do you do?” may have a handler.

The Champagne case

The training also reportedly includes a story about a US diplomat in Russia. An espionage operation first sent a blonde woman to his hotel room with Champagne. He said no, so they sent a redheaded woman. Then they sent a young man. NDTV summed up the lesson: they’re relentless, and they will keep trying until they find out what you like.

The rules

The Journal says employees are told to:

• avoid wearing OpenAI-branded clothing outside the office,
• use privacy screens on public transport,
• wipe whiteboards after meetings, in case kitchen or cleaning staff are compromised,
• and watch out for people listening in at cafés.

Staff get frequent reminders, including on Slack, which is quite something for a company that might like its own chatbot to handle that.

Why it’s not just paranoia

The warnings got sharper this week. Earlier this month the FBI arrested an OpenAI employee accused of planting a recording device in a company conference room. He has been charged and has not been convicted, and the case is still going. With AI labs heading for trillion-dollar valuations and hiring fast, security experts say they’ve become prime targets.

The Snark take

For years, tech workers’ worst workplace danger was a bad standup. Now it’s a stranger with a toddler and too much interest in their job. Our advice to OpenAI staff: if someone at a bar laughs at your joke about attention mechanisms, be suspicious. Nobody laughs at that.