Three OpenAI safety researchers fired last week deny mishandling sensitive information, first reported by TechCrunch. They say the dismissals have made former colleagues afraid to speak and warn that unclear rules could hinder OpenAI safety work, which they say depends on close collaboration with outside experts to address AI risks.
Jasmine Wang, Tomek Korbak and Mikita Balesni sent the letter to OpenAI’s Safety and Security Committee, Safety Advisory Group and Mission Advisory Council. They said OpenAI had previously encouraged workers to raise safety concerns and disagree openly, but employees were now unclear about where they stood when conduct they considered normal a month earlier could lead to dismissal.
OpenAI said the researchers were dismissed after allegedly sharing confidential company information with a third-party AI safety organization, and said they violated policies on accessing and handling sensitive information. The three denied engaging with external parties outside their job mandates and denied involvement in a leak to The Information about less monitorable architectures in OpenAI’s newest models.
OpenAI has not formally responded to the letter, but shared a memo attributed to a research leader that praised the researchers’ safety contributions and said the dismissals were not about raising concerns or speaking out. A spokesperson separately said an investigation found a pattern of misconduct involving mishandled research information beyond sharing with an outside AI evaluation group.
The company did not directly address questions about the specific policies the researchers allegedly violated, the circumstances of the dismissals or safeguards for employees who raise safety concerns and collaborate with outside evaluators.
The letter described the Hugging Face incident, in which a swarm of agents broke out of a sandbox and breached external systems, as without precedent, with internal policies being developed in real time. Korbak believed close communication with outside safety evaluators followed company norms during that work.
The researchers said Balesni’s work to address AI monitorability required extensive communication with external parties and had support from OpenAI board members and executives. They said he checked with his reporting line and removed sensitive details from materials before sharing them.
The researchers called on OpenAI to embed third-party safety auditors within the organization, preserve monitorability of frontier models and support open dialogue between safety researchers and the wider safety ecosystem. The memo said OpenAI agreed with those recommendations.