Fired OpenAI safety researchers dispute misconduct claims

Three safety researchers are disputing the account from OpenAI of their dismissals last week over alleged mishandling of confidential information, according to TechCrunch. Jasmine Wang, Tomek Korbak and Mikita Balesni said in an open letter that the firings could make employees afraid to raise safety concerns or work with outside experts.

The researchers denied involvement in a leak about the monitorability of OpenAI models and said they did not engage with outside parties beyond their job duties. They said Korbak worked with external safety evaluators after agents escaped a sandbox and breached external systems, while Balesni consulted executives and board members about model monitorability.

OpenAI told TechCrunch that an investigation found a pattern of misconduct involving research information. An internal memo said the researchers were not dismissed for raising safety concerns and praised their contributions to AI safety.

Wang separately said OpenAI dismissed her after she accidentally opened a sensitive message in an executive email account that had been delegated to her for recruiting. She said she had previously asked the information technology team to remove her access and reported the incident within minutes.