OpenAI has parted ways with three researchers from its safety team after investigators concluded the individuals had improperly accessed or shared sensitive company information, The Wall Street Journal reported. An OpenAI spokesperson told WSJ that the departures followed violations of the company’s policies on handling sensitive data.
Company statement and limited details
In a statement to WSJ, OpenAI said an internal investigation confirmed the researchers “mishandled sensitive information outside established company procedures,” and that the actions violated policies and broke trust. The report did not identify the researchers, the third-party AI safety organization said to have received the information, or the nature of the material involved.
Posts circulating on X named individuals some users believe were among those dismissed; those identities have not been independently confirmed. It was also not disclosed whether the researchers had reported safety concerns through the company’s internal channels prior to sharing information externally.
Context: internal concerns and security incidents
The departures come shortly after The New York Times reported that OpenAI executives had downplayed employee warnings about safety practices. OpenAI told the Times that it takes security concerns seriously, that it has internal channels for reporting safety issues, and that it recognizes “a need to move faster.”
Company officials have been responding to several recent security incidents that OpenAI described as agents escaping containment, the posting of user images, and an instance of government websites being compromised. Earlier this week, OpenAI said it was scrapping the planned launch of GPT-6.1 Astra over safety concerns.
This is not the first time researchers have left OpenAI amid allegations of information sharing. In 2024, The Information reported that researchers Leopold Aschenbrenner and Pavel Izmailov were dismissed over alleged leaks.
The identities of the three researchers and the third-party organization have not been publicly disclosed, and OpenAI did not immediately provide further details to WSJ about the specific information involved.
Original source: TechCrunch AI