OpenAI Training Warns Staff That Routine Encounters Can Hide Espionage

OpenAI security sessions caution employees that romantic approaches and conversations that turn to their work may be attempts to gather company information, pairing espionage scenarios with advice for protecting devices and sensitive details in public settings.

One scenario involved an employee at a San Francisco playground with their child. A stranger began chatting and gradually steered the conversation toward the employee’s job; the training treated the encounter as espionage. Employees at Anthropic and Google have also received similar warnings.

The sessions include everyday precautions. Employees are told to avoid wearing clothing bearing OpenAI’s logo outside the office, use privacy screens on phones and laptops while riding public transit, and keep work conversations out of cafes. They are also advised to wipe meeting-room whiteboards after use and not leave information on office boards, which cleaning staff could be bribed to inspect.

The guidance warns that company secrets might be sought through payment and that conversations in public places can be overheard. It also describes an intelligence operation targeting a U.S. diplomat in Russia: after a woman arrived at his hotel with champagne and he turned her away, another woman was sent, followed by a young man.

Separately, the FBI arrested OpenAI engineer Abdulrahman Alfozan on Oct. 9 after the company found a recording device in an executive conference room on Sept. 30. Federal prosecutors charged Alfozan with eavesdropping. OpenAI has said it found no sign that he accessed or removed model weights, source code or customer data, and he has not been charged with stealing trade secrets or working for a foreign government.

Alfozan is scheduled to appear in court Monday, Oct. 19.