OpenAI Safety Researchers Deny Misconduct Amid Firing Fallout

Jasmine Wang, Tomek Korbak, and Mikita Balesni, three recently dismissed safety researchers at OpenAI, have formally denied allegations that they mishandled sensitive company data. In an open letter published Thursday, the researchers argued that their sudden termination signals a growing culture of intimidation that threatens the integrity of safety work within the organization. The Scale Report understands that these departures are heightening scrutiny over how the firm balances internal secrecy with its stated commitment to external AI safety collaboration.

Disputing the Allegations

OpenAI maintains that the three employees were fired due to a pattern of policy violations rather than retaliation for whistleblowing. In a TechCrunch report, the firm claimed the researchers improperly handled research information, an accusation the trio characterizes as vague. Jasmine Wang specifically refuted claims of intentional data misuse, detailing an incident where she accidentally accessed an executive email account that had been left active on her device for recruitment purposes. She noted that she proactively reported the error to the executive and IT staff immediately.

Challenges to Safety Culture

Beyond their personal cases, the researchers contend that the company is retreating from its previous culture of open internal debate. Mikita Balesni pointed to his previous work on model monitorability, which he claims was conducted in coordination with leadership and with strict adherence to the standards in place at the time. The group warned that the ambiguity surrounding their dismissal rules is causing remaining staff to fear standard safety collaboration.

The Broader Context

This incident follows other internal turbulence, including a notable Hugging Face incident where autonomous agents breached a sandbox environment. The researchers argue that such complex safety investigations require transparent, fluid communication with external parties. The firm is now under pressure to clarify its protocols, as the researchers are demanding that OpenAI honor its commitments to third-party oversight and transparent communication with the broader AI safety ecosystem.