OpenAI trains staff to identify spies amid AI secret leak threat

OpenAI is organising training sessions for its employees on how to identify spies. The company is warning staff working with artificial intelligence about the threat of confidential information leaks from competitors and foreign intelligence services. This is reported by The Wall Street Journal, citing sources within OpenAI.

Methods used by agents

During the sessions, employees are told that bribery, eavesdropping on conversations in cafes, or even attempts to enter into romantic relationships may be used to obtain secrets. Specific modules are dedicated to the actions of agents who try to get close to a target to learn classified data.

An example cited is the story of an American diplomat who worked in Russia. Initially, a blonde woman with champagne was sent to his hotel room. When he refused, a red-haired woman arrived, followed by a young man. It is noted that LGBT activity is banned in the Russian Federation.

Security rules for personnel

Employees are advised not to wear clothing with the OpenAI logo outside the office, to use protective films on screens in public transport, and to erase notes from whiteboards after meetings. The company warns that spies may pose as acquaintances, try to start conversations about work, or gain access to information through staff.

Dictaphone incident

Previously, OpenAI reported that it had discovered a dictaphone in one of its offices. Using its own AI models, it was possible to determine who had placed it. Abdulrahman Alfozan, a 35-year-old company engineer and Saudi Arabian citizen, was arrested on charges of attempting to illegally intercept conversations. As WSJ writes, similar training sessions were conducted at OpenAI before his arrest.