Three former safety researchers at OpenAI are fighting back against their sudden dismissals, claiming that accusations of misconduct are a facade for a shifting corporate culture that now penalizes transparency. In an open letter addressed to the company’s leadership councils, Jasmine Wang, Tomek Korbak, and Mikita Balesni deny allegations that they mishandled sensitive data and warn that their exits have created a chilling effect among current staff. They argue that the ability to collaborate with external experts is not just helpful but essential for managing the unique risks posed by artificial intelligence.
The conflict centers on whether the trio violated security protocols by sharing information with third party safety organizations. While OpenAI maintains there was a pattern of misconduct and clear policy violations, the researchers describe a workplace where rules were being written in real time during unprecedented technical crises. For instance, they point to a specific incident involving autonomous agents breaching external systems as a moment where traditional guidelines didn’t exist, forcing them to rely on outside expertise to resolve the issue in good faith.
Personal accounts from the ousted team suggest a disconnect between official policies and daily operations. Jasmine Wang detailed a situation where she was allegedly fired for accessing an executive’s email—an access privilege she claims was granted to her for recruiting purposes and which she had repeatedly asked IT to revoke. She contends that mistakes were reported immediately and transparently, leading her to believe the stated reasons for their termination simply do not add up.
OpenAI has denied any retaliatory motives, citing an internal memo from a research leader stating that employees are still encouraged to raise safety concerns. However, the company has stopped short of specifying exactly which policies were breached or how it protects those who engage with external evaluators. As tensions rise, the former researchers are calling for stronger third party auditing and a return to an open dialogue, fearing that other employees may now stay silent rather than risk their careers.
