The dispute began after OpenAI dismissed Jasmine Wang, Tomek Korbak and Mikita Balesni from its safety team following an investigation into the handling of sensitive information. The company said the three had violated policies, while the researchers’ work had intersected with the investigation of a rogue-agent incident involving Hugging Face.¹

On Oct. 8, the former employees answered with a joint open letter, denying that they had improperly shared information with METR, the outside AI-safety group involved in reviewing the Hugging Face episode. They argued that an unprecedented incident had forced internal procedures to evolve in real time—and that close contact with external evaluators was part of doing safety work responsibly.²

Their sharper concern was cultural. “Terminations such as ours, executed and communicated so abruptly, are chilling the open culture OpenAI has prized in the past,” they wrote.² Wang said the practical message to remaining staff was bleak: “Raise concerns or work closely with outside safety groups, and you could be next.”¹ The trio urged OpenAI to preserve model monitorability, maintain meaningful access for third-party auditors and clarify protections for employees who raise alarms.

OpenAI rejects the retaliation charge. In an internal memo shared publicly after the letter, a research leader said: “These decisions were not about raising safety concerns or speaking out. We have always encouraged that and always will.”³ The company says a thorough investigation found a broader “pattern of misconduct” and a significant breach of trust, without publicly detailing the conduct at issue.¹

That leaves the core disagreement unresolved. The researchers see opaque dismissals as a threat to independent scrutiny precisely as advanced systems become harder to monitor; OpenAI says accountability for sensitive information is compatible with, not contrary to, an open safety culture. It has said it still intends to work with third-party safety assessors and is finalizing related contracts.¹