Leading AI companies are regularly briefing employees on how not to become a source of leaks, according to a Wall Street Journal report. OpenAI warns its teams about bribery, eavesdropping in cafes, cleaning crews that may have been compromised, and honey traps built around seduction; Anthropic and Google staff have received similar briefings.

The OpenAI rules are concrete: outside the office, hide anything carrying the company logo, and treat kitchen and cleaning staff as possible leak channels. Reminders arrive on a schedule, including through Slack.

The paranoia has a price tag behind it. A handful of labs hold a concentration of value, several have large IPOs approaching, and the WSJ puts the valuations discussed at roughly $3.5 trillion. Model weights, training data and architectural decisions are now valuable enough to attract the classic methods of espionage.

Related: Apple Sues OpenAI for Trade Secret Theft in Blockbuster Case, Anthropic Blocks Claude Code Access in China Over Model Distillation

WSJ: AI companies play cat and mouse with secret-stealing spies