Fired OpenAI safety staff deny leaking, defend Astra monitoring

The dispute centers on two sensitive areas: OpenAI's ability to monitor its advanced models, and its engagement with outside safety organizations such as METR. Korbak served as technical point of contact for METR, the third-party research firm OpenAI asked to conduct an independent audit of the Hugging Face incident. Balesni and Korbak worked on OpenAI's investigation of that incident.

The letter also responds to internal versions of events. The former employees say ongoing internal and external communications are making current employees afraid to speak up, and they deny leaking concerns about monitoring OpenAI's latest Astra model to The Information, which published a Sept. 1 article about it.

OpenAI has not disclosed specific reasons for the firings and did not link them to METR communications or the Hugging Face incident. The company told Fortune it found a pattern of misconduct, including violations of its information-handling policies. Korbak said OpenAI told him verbally he was fired over how he communicated with METR. Wang said she was told she accessed an unnamed executive's email, which she says the company had given her for recruiting and that IT did not remove when she asked. Balesni said the researchers were fired for prioritizing safety over OpenAI's near-term corporate interests.