Why OpenAI Safety Researchers Were Fired
OpenAI safety researchers Jasmine Wang, Tomek Korbak, and Mikita Balesni publicly denied the company’s claim that they mishandled research information. They said their contacts with outside safety groups were intended to support evaluation, monitoring, and risk research, and that they believed the work followed the norms in place at the time.
OpenAI said an investigation found a broader “pattern of misconduct,” extending beyond the sharing of information with an external AI evaluation group. An internal memo said the dismissals were not retaliation for raising safety concerns or speaking openly. However, the company has not publicly identified the specific policies involved or detailed how the terminations were handled.
The Dispute Exposes AI Safety Governance Tensions
The researchers warned that abrupt dismissals and unclear rules could discourage employees from reporting risks and weaken third-party oversight. They also denied involvement in a media leak about model monitorability, and said that while handling an unusual agent incident and related research, they communicated with internal leaders and external evaluators.
The dispute is therefore about more than an employment decision. It highlights how frontier AI labs must balance confidentiality with outside scrutiny and safety collaboration. Researchers who must guess which discussions could trigger discipline may become less willing to investigate risks, while clear and enforceable controls for sensitive information remain necessary.
The Key Questions That Remain
The two accounts are still unresolved. OpenAI has not published the specific policy basis, investigative evidence, or protections available to employees who raise safety concerns. The researchers have urged the company to support third-party safety audits, preserve frontier-model monitorability, and maintain open dialogue. Further transparency will show whether this was an isolated dispute or evidence of a broader governance shift.