OpenAI defends firing three AI safety researchers over breach of trust

OpenAI has defended its decision to dismiss three AI safety researchers, Jasmine Wang, Tomek Korbak, and Mikita Balesni, following a dispute over the handling of confidential company information.

The company rejected claims that the dismissals were linked to the researchers’ work on AI safety.

OpenAI said an internal investigation found a serious breach of trust and concluded that the three researchers had violated its rules for accessing and handling sensitive information. However, the company has not publicly disclosed all the details of the violations.

The researchers have disputed OpenAI’s account, saying they acted in good faith while working with external AI safety experts. In an open letter, they warned that their dismissals could discourage employees from raising concerns about the risks posed by artificial intelligence.

They argued that collaboration with outside experts is important for identifying weaknesses in advanced AI systems. The researchers also denied being the source of a report by The Information that raised concerns about OpenAI’s Astra model and the challenges of monitoring the behaviour of advanced AI systems.

Wang separately claimed she was dismissed after accidentally opening a sensitive email in an executive’s inbox. She said she had previously been granted access for recruiting work and had asked the IT team to remove it. OpenAI has not publicly confirmed all the details of her account.

The company has maintained that internal discussions about research and safety are a regular and important part of its work, and denied punishing employees for raising concerns.