Fired OpenAI Safety Researchers Reject Misconduct Claims, Warn of Climate of Fear
Three recently dismissed OpenAI safety researchers say in an open letter that they did not breach company procedures, and warn that their firing discourages open safety work. OpenAI says an investigation found a pattern of mishandling research information.
Jasmine Wang, Tomek Korbak and Mikita Balesni, three safety researchers dismissed by OpenAI last week, published an open letter on Thursday. In it, they reject the company's claims that they handled sensitive information in violation of established procedures.
The letter is addressed to OpenAI's Safety and Security Committee, Safety Advisory Group and Mission Advisory Council. The researchers write that communications around the firings have left former colleagues afraid to speak and work in ways that, until last week, were a normal part of life at the company.
The company's position
OpenAI said the three violated policy by "accessing and handling sensitive company information." A spokesperson told TechCrunch that an investigation revealed a "pattern of misconduct" going beyond sharing information with an outside AI evaluation group. The company did not say which specific policies were breached, nor did it address the circumstances of the dismissals or how it protects employees who raise safety concerns. TechCrunch was also given an internal memo from a research leader, which stated that the decisions were not about raising safety concerns and that employees are not terminated for doing so.
The researchers' account
In the letter, the three deny involvement in a leak to The Information about architectures that make chain-of-thought reasoning harder to monitor. They also deny engaging with outside parties beyond their job mandates.
According to the letter, Korbak believed that close contact with outside evaluators during the investigation of the Hugging Face incident, in which a swarm of agents escaped its sandbox and breached external systems, fell within company norms. Balesni worked on the monitorability problem, kept his reporting line informed and, per the letter, had support from board members and executives. On X, Wang said she was fired over access to an executive's email that the company had delegated to her for recruiting, and that she reported opening a sensitive message by mistake within minutes.
The researchers urge OpenAI to keep supporting open dialogue with the wider safety ecosystem and to preserve the monitorability of frontier models.
