OpenAI briefs staff on honeypots and playground approaches as AI labs harden against spies

What's going on? The Wall Street Journal reported on 14–15 October 2026 that OpenAI has been training employees on how foreign or state-sponsored spies and competitors might try to extract secrets — including romantic "honeypot" approaches, cash offers, café eavesdropping, and a scenario in which a stranger repeatedly shows up with a child at a San Francisco playground used by an OpenAI employee and steers conversation toward their work.

Operational advice (as reported): Staff have also been urged not to wear company-branded clothing outside the office, to use privacy screens, and to wipe whiteboards after meetings in case cleaners or other onsite workers are recruited. The Journal said employees at Anthropic and Google have received similar espionage warnings.

Context, not the same story: These briefings were already under way before federal authorities arrested OpenAI engineer Abdulrahman Alfozan over an alleged hidden recorder in an executive conference room (Dealroom note on that arrest: separate). Treat the playground scenario as a training example described by the Journal, not as a newly proven public incident with named defendants beyond that arrest.

Why it matters: Frontier AI labs are starting to look operationally more like defence contractors — with insider-threat playbooks and family-adjacent social engineering on the risk list. That is a sticky culture and security signal for how the industry is professionalising under competitive and geopolitical pressure.

Sources: Wall Street Journal, "AI Companies Play Cat-and-Mouse With Secret-Stealing Spies" (14 Oct 2026 evening ET), as reported by AI Affairs and India Today (15–16 Oct 2026).