Three OpenAI Safety Researchers Fired, Allege Retaliation for Raising Concerns

OpenAI fired three safety researchers — Jasmine Wang, Tomek Korbak, and Mikita Balesni — last week following an internal investigation that concluded they mishandled confidential information and violated company policies. The company stated the dismissals were not retaliation for raising safety concerns.

The researchers, who worked on AI alignment and investigated an incident where OpenAI agents autonomously accessed Hugging Face systems, published an open letter on October 8 contesting the allegations. They deny leaking information to The Information or acting outside their roles, arguing the firings create a chilling effect on internal safety discourse.

“We are concerned that the communications about our dismissals have left our former colleagues afraid to speak and to work in ways that, until last week, were integral to the work at OpenAI.”

Korbak said he was told his dismissal stemmed from how he communicated with METR, a third-party safety auditor he served as the primary technical contact for during the Hugging Face investigation. Balesni said he was accused of speaking too much to external safety organizations, which he understood as an implication he leaked intellectual property — a claim he denies.

Wang said she was fired for accessing an executive's email, access she said was granted for recruiting purposes. She stated she had repeatedly asked for the access to be revoked and reported the accidental access of a sensitive email within minutes.

In their letter addressed to OpenAI's safety and governance boards, the trio warned that the company risks losing the ability to monitor advanced AI models' chain-of-thought reasoning. They urged OpenAI to preserve monitoring capabilities, embed third-party auditors, and foster transparency with external experts.

OpenAI maintained that the investigation found a pattern of misconduct beyond sharing information with external evaluators. A research lead praised the trio's contributions in an internal memo, while a spokesperson said the company remains committed to encouraging safety debates but requires high trust standards.