The dispute began publicly after the Wall Street Journal reported the dismissals on Oct. 1. OpenAI said the three employees had mishandled sensitive information outside established procedures, breaching the trust needed for its work. The company has not publicly detailed the alleged conduct, and did not tie the firings to METR, the outside safety group involved in reviewing the Hugging Face incident.¹

For Tomek Korbak, Jasmine Wang and Mikita Balesni, that lack of detail is central to the case. In an Oct. 8 open letter, they denied leaking concerns about the monitorability of OpenAI’s Astra model and rejected claims of improper contact with METR. They argued that their abrupt exits were already changing behavior inside the company: “Terminations such as ours, executed and communicated so abruptly, are chilling the open culture OpenAI has prized in the past.”²

The researchers cast the conflict as more than an employment dispute. Korbak, who served as a technical contact for METR during the unprecedented Hugging Face investigation, said communication with the evaluator was part of his job. Wang said an executive-email access issue stemmed from recruiting permissions she had asked IT to remove. Balesni said his work on model monitorability involved internal coordination and careful removal of sensitive details before anything was shared.²

Their warning is that unclear rules could deter the very people tasked with surfacing danger. “You can’t build AGI safely if the people closest to the risks are afraid to speak,” Wang said.² They called on OpenAI to preserve monitorability, embed third-party auditors and restore a transparent culture of disagreement.

OpenAI’s response draws a sharper line: debate is protected, but information rules are not optional. In an internal memo shared publicly Friday, a research leader said, “These decisions were not about raising safety concerns or speaking out,” adding that an investigation found policy violations involving sensitive information and a broader breach of trust.³ The company says it remains committed to outside safety assessors and agrees on the importance of monitoring frontier models. The unresolved question is whether employees believe those assurances when the boundaries of permitted safety work remain undisclosed.