OpenAI warns staff that spies may use romance to steal secrets

OpenAI has warned its workforce that intelligence operatives and other adversaries may use romantic overtures as a vector for social-engineering attacks aimed at stealing sensitive research and proprietary business information. The internal alert emphasizes that security awareness must extend beyond the protection of devices and networks to encompass personal interactions during travel, recruiting activities, and personal relationships.

The guidance highlights that employees should be vigilant when encountering unusual or high-pressure requests for access, regardless of how those requests are framed socially. The company is reinforcing that the human element remains a critical attack surface, and that cultivating relationships can be a deliberate tactic to bypass technical safeguards and extract confidential data.

For leaders in strategy, human resources, legal, and communications, the warning underscores the need to integrate these broader social-engineering risks into security training and insider-threat programs. The advisory serves as a reminder that operational security protocols must account for the full spectrum of interpersonal engagement, not just conventional digital hygiene.