The dispute began after OpenAI fired Jasmine Wang, Tomek Korbak and Mikita Balesni, three researchers working on AI safety, following an internal investigation. The company said the decision stemmed from a “pattern of misconduct” and violations of policies governing sensitive research information—not from the researchers’ safety work or criticism.¹

The researchers answered with an open letter, saying the episode had left former colleagues afraid to communicate and operate in ways that had recently been routine. They argued that safety staff need latitude to work with outside experts, particularly when confronting emerging risks, and warned that abrupt dismissals and vague standards could turn that collaboration into a career hazard. “Terminations such as ours, executed and communicated so abruptly, are chilling the open culture OpenAI has prized in the past,” they wrote.¹

Their account centers on the blurred rules around extraordinary safety incidents. Korbak said external contact during the investigation into the Hugging Face breach was consistent with OpenAI norms, while Balesni said he coordinated internally and stripped sensitive details before sharing material. Wang, meanwhile, said she accidentally opened a sensitive executive email through access originally granted for recruiting, reported it within minutes and had previously asked IT to remove the access.¹

On Friday, OpenAI doubled down. It called the episode a “significant breach of trust,” said a thorough inquiry had found clear policy violations, and insisted the dismissals “were not about raising safety concerns or speaking out.”² The company also said it shares the researchers’ goal of preserving frontier-model monitorability.

That overlap is the dispute’s sharpest irony: both sides say safety requires openness and external scrutiny. The researchers contend the firings make those safeguards harder to practice; OpenAI says rules on sensitive information are precisely what make responsible safety work possible.