OpenAI fires three safety researchers for "mishandling research information"

Three OpenAI safety researchers - Jasmine Wang, Tomek Korbak, and Mikita Balesni - were fired last week for allegedly mishandling sensitive research information and sharing it with an outside AI safety group. They published an open letter denying any misconduct, saying their actions followed past norms and were necessary for external collaboration on monitorability and safety. They argue their dismissals signal a chilling effect that will stifle internal debate and third‑party accountability, noting that work on monitorability and the response to a Hugging Face incident involving rogue agents required close external coordination. Korbak says he believed his outreach complied with evolving internal rules, Balesni says he coordinated with board members and removed sensitive details before sharing, and Wang says she was dismissed after accidentally opening an executive’s email she had been granted access to for recruiting and promptly reported it.

OpenAI has not publicly detailed which policies were violated; an internal memo praised the researchers’ contributions and denied retaliation while a spokesperson said an investigation found a “pattern of misconduct” beyond sharing with an evaluation group. The firings have intensified scrutiny amid other safety incidents and leaks, and the researchers call on the company to reaffirm commitments to embedded third‑party auditors, preserve model monitorability, and protect a culture where safety staff can speak and collaborate without fear.