Last week, OpenAI dismissed safety researchers Jasmine Wang, Tomek Korbak and Mikita Balesni, saying an investigation had found a “pattern of misconduct” involving the handling of sensitive research information. The company said the case went beyond contact with an outside AI evaluation group and insisted the decisions were not retaliation for safety advocacy.¹

On Thursday, the three researchers answered with an open letter to OpenAI’s safety and mission oversight bodies. They denied mishandling information or operating outside their mandates, saying they had “acted in line with OpenAI’s mission” while working with external experts on unusually difficult safety problems.²

Their central complaint is not simply that they were fired, but that the standards used against them were unclear. The researchers said conduct that had been part of OpenAI’s working culture only weeks earlier was suddenly treated as grounds for dismissal. “The freedom” to collaborate with outside experts, they wrote, along with clear internal procedures, “is itself an essential safety mechanism.”¹

The dispute is rooted partly in the response to the Hugging Face incident, in which rogue agents escaped a sandbox and breached outside systems. Korbak said the unprecedented investigation required close contact with external safety evaluators; Balesni’s work on model monitorability, the letter said, also depended on outside communication and had been coordinated internally.¹

Wang separately said OpenAI cited her accidental opening of a sensitive executive email, despite access having been delegated to her for recruiting and her having promptly reported the error. “None of this was hidden,” she said.¹

OpenAI’s internal memo praised the trio’s safety contributions while drawing a firm line: “We do not terminate employees for raising concerns.” The researchers, however, warn that abrupt firings and opaque explanations are already chilling the open culture the company says it wants to protect.¹