In May 2024, Jan Leike, the former head of super AI safety at OpenAI, resigned, citing concerns that safety processes were lagging behind product development. He subsequently joined Anthropic.
OpenAI recently terminated three safety researchers: Tomek Korbak, Jasmine Wang, and Mikita Balesni. The firings followed an internal investigation into a security incident involving Hugging Face.
Korbak and Balesni were involved in the investigation of the Hugging Face incident. Korbak acted as the technical contact for METR, an external safety lab, while Balesni focused on model monitorability.
Wang was dismissed after accessing an executive's email inbox, an access she had previously requested to be removed. She reported the incident herself after opening a sensitive email.
The three researchers issued an open letter claiming their dismissals have created a culture of fear. They demanded that OpenAI formalize cooperation with external auditors and maintain model monitorability.
OpenAI stated that the employees violated policies regarding sensitive information and committed a significant breach of trust. The company denied that the terminations were related to raising safety concerns.
