OpenAI responded on October 9, 2026, to a letter published the previous day by former safety and alignment employees Tomek Korbak, Jasmine Wang, and Mikita Balesni. The company said an internal investigation found policy violations involving access to and handling of sensitive information, along with a significant breach of trust. It denied that the dismissals were about raising safety concerns or speaking out.
OpenAI responds to the former employees’ letter
The letter, published October 8, was addressed to OpenAI’s safety oversight groups. The three former employees said they had acted within their work mandates and argued that the dismissals and uncertainty around them could discourage open discussion and collaboration with outside safety organizations.
What OpenAI says—and what the former employees say
OpenAI’s explanation centers on its investigation into the handling of sensitive company information. The former employees dispute the suggestion that their outside work fell beyond their roles. Their letter also describes policies for the unprecedented Hugging Face investigation as developing in real time.
The company said the dismissals were not about raising safety concerns or speaking out. The former employees, meanwhile, urged OpenAI to maintain open dialogue and clear procedures for staff working with external safety organizations.
Why METR and external oversight matter
METR, short for Model Evaluation and Threat Research, is an external evaluator that OpenAI asked to assess the July 2026 Hugging Face incident. The joint letter identifies Korbak as the technical point of contact for METR during that work.
The former employees called for continued external safety audits and the ability to monitor frontier models—that is, to track how advanced systems behave. They argue that access for outside evaluators and collaboration across organizations are important parts of safety work.
The three employees’ individual accounts
Korbak said he was told verbally that his dismissal concerned how he communicated with METR. Wang’s account was that access to an executive’s email had been delegated for recruiting; she said she asked IT to remove it when it was no longer needed and reported accidentally opening a sensitive email.
Balesni denied sharing OpenAI intellectual property and said his work had been coordinated with his reporting line, research leadership, and the board.
OpenAI’s stated plan for outside assessors
In its October 9 statement, OpenAI said it was finalizing contracts with third-party safety assessors and intended to announce details in the coming weeks.
