OpenAI has incorporated counter-espionage material into its internal security training as part of a broader effort to prevent leaks of sensitive business information.
The Wall Street Journal reported that the company used a spy story in one training session. In that example, a foreign intelligence agency allegedly tried to seduce a U.S. diplomat stationed in Russia. A blonde woman carrying champagne first knocked on the diplomat’s hotel room door. After being turned away, a red-haired woman was sent, followed later by a young man.
OpenAI used the case to warn employees that spies may try to get close to targets through romance or intimate relationships in order to obtain confidential information.
Training also covers casual social contact
The report said OpenAI has also told employees to be careful with seemingly ordinary conversation in daily life.
In another training example, a stranger frequently showed up at a playground in San Francisco with a child, gradually became familiar with an OpenAI employee, and then started asking about the person’s work during casual chats. The company used the scenario to show how repeated contact in informal settings can be used to gather information.
Staff were told to reduce exposure of sensitive material
Beyond case studies, OpenAI has reportedly instructed employees not to wear clothing with the OpenAI logo outside when possible, to place privacy filters on their computer screens when using laptops on the subway, and to wipe whiteboards clean after meetings so cleaning staff do not see sensitive information.
According to the report, the training had already been underway before this week’s bugging case became public.
DOJ disclosed bugging case on Oct. 13
On Oct. 13, the U.S. Department of Justice disclosed that an OpenAI engineer was suspected of installing a recording device in an executive meeting room and was arrested by the FBI.
Anthropic and Google are also carrying out similar counter-espionage training.