OpenAI has disclosed that it terminated three employees who worked on safety-related matters, stating they mishandled sensitive information in violation of company rules.

People familiar with the matter said the incident involved employees leaking or mishandling sensitive internal materials to an outside organization that conducts AI evaluations. An OpenAI spokesperson said in a statement: "We terminated our employment relationship with three employees for violating company rules and improperly accessing and handling sensitive company information. An investigation confirmed that these individuals failed to follow established procedures for handling sensitive information externally, violating internal policies and undermining the trust necessary to carry out our work."

The dismissals come amid mounting concerns about the risks posed by advanced AI models. OpenAI and other AI companies have long collaborated with various evaluation organizations. In response to growing risk concerns, major companies have recently said they are willing to let AI safety agency personnel embed within their organizations to audit disclosures related to technological risks. But such cooperative arrangements typically impose strict constraints on what outside organizations can access and publicly disclose.

An OpenAI spokesperson said that among those dismissed, one focused on AI safety, one researched AI alignment, and the third was a research project manager in the AI alignment area. The spokesperson added that the investigation found a series of improper actions by the group, including mishandling research materials in violation of company rules.