Three dismissed OpenAI safety researchers have published an open letter challenging the company’s account of their conduct. Jasmine Wang, Tomek Korbak and Mikita Balesni warn that unclear rules could discourage safety work and outside collaboration. OpenAI has responded that its investigation found misconduct involving sensitive information, while denying retaliation for raising safety concerns. The opposing accounts remain disputed.
Jasmine Wang, Tomek Korbak and Mikita Balesni, three researchers dismissed from OpenAI’s safety work last week, published an open letter on Thursday. They challenge the company’s description of their conduct and say they acted within its mission and working norms. The letter brings their own account into the dispute surrounding their dismissals.[1], [2]
The researchers deny involvement in a leak about model architectures whose reasoning is harder to monitor. Wang separately says that executive-email access had been delegated to her for recruitment. She says she reported accidentally opening a sensitive message. These are the researchers’ statements about the conduct at issue.[1]
They also warn that unclear rules and abrupt dismissals could discourage safety staff from speaking openly and collaborating with external evaluators. Their concern links the handling of this personnel dispute to how safety work is conducted inside the company.[1]
OpenAI responded on Friday that its investigation found a significant breach of trust and misconduct involving sensitive research information. It says clear policies were broken and denies that the dismissals were retaliation for speaking out or raising AI safety concerns. The company says the investigation uncovered matters beyond those described in the letter, without publicly detailing them. The competing accounts remain disputed.[1], [2]
