Fired OpenAI safety researchers deny misconduct, warn of chilling effect

Three former OpenAI safety researchers, Jasmine Wang, Tomek Korbak, and Mikita Balesni, have publicly contested their dismissal last week, denying allegations of policy violations and warning that their termination threatens the company foundational safety culture. In an open letter addressed to OpenAI safety governance bodies, the researchers rejected claims that they improperly shared confidential information, emphasizing that collaboration with external safety experts remains critical to addressing advanced AI risks. They argued that the abrupt firings generate a chilling effect, leaving remaining employees uncertain about acceptable conduct and potentially stifling vital safety oversight. OpenAI justified the terminations following an internal investigation that uncovered a pattern of misconduct regarding the mishandling of research data, explicitly stating that the decisions were not retaliatory for raising safety concerns. Company leadership praised the researchers technical contributions while reiterating strict adherence to established data protocols. The dispute has intensified scrutiny over OpenAI internal safety frameworks, particularly following recent incidents involving unmonitored model behaviors and the Hugging Face agent breach. The former researchers noted that during the latter investigation, internal policies were still being formalized, and their external communications with safety evaluators were conducted in good faith to ensure model monitorability and build industry trust. Wang further clarified her individual termination, alleging she was dismissed after accidentally accessing an executive email that her IT access had not revoked despite her repeated requests to remove it. She emphasized the immediate disclosure of the incident and questioned the consistency of the grounds applied to all three cases. The researchers urged OpenAI to honor its public commitments to integrate independent third-party auditors, maintain rigorous model transparency, and preserve an environment where safety personnel can communicate freely with external experts. OpenAI leadership acknowledged these recommendations in internal communications, though company officials declined to specify exact policy breaches or outline formal protections for safety collaboration. The incident has reignited industry debate regarding the balance between corporate data security and the necessity of open, external accountability in frontier AI development.