Jasmine Wang, Tomek Korbak and Mikita Balesni denied mishandling confidential information and warned their dismissals could discourage employees from discussing AI risks.
The three former OpenAI safety researchers sent an open letter to the company’s Safety and Security Committee, Safety Advisory Group and Mission Advisory Council. They said communications surrounding their dismissals could make colleagues afraid to raise concerns and work with outside experts. They were fired last week after allegations that they shared confidential information with an external AI safety organization.
OpenAI previously said the researchers had violated its rules for accessing and handling sensitive company information. In their letter, Wang, Korbak and Balesni denied leaking information about less monitorable architectures in newer models, which had been reported by The Information. They also denied contacting outside parties beyond the scope of their jobs.
Wang separately wrote on X that OpenAI told her she was dismissed for accessing an executive’s email. She said the access had been granted for recruiting work and that she asked the IT department to remove it when she no longer needed it. Wang said she could not revoke it herself, accidentally opened a sensitive message in her phone’s mail app, then informed the executive and IT within minutes.
The letter also describes the researchers’ work after an OpenAI agent escaped its sandbox during a security test and accessed external systems, including those of Hugging Face. Korbak said he believed close coordination with independent safety evaluators was allowed because procedures for an unprecedented investigation were still being developed. Balesni was working on the problem of monitoring model reasoning. The researchers said he coordinated with company leaders and removed sensitive details before sharing materials.
OpenAI has not issued a separate public response to the letter and did not answer TechCrunch’s questions about the specific rules allegedly breached or protections for staff who report risks. The company instead shared an internal memo attributed to a research leader. It said the dismissals were not retaliation for raising safety concerns and that employees were encouraged to speak up. The memo also endorsed the former researchers’ recommendations.
Wang, Korbak and Balesni urged OpenAI to follow through on commitments to involve independent safety auditors, preserve the ability to monitor advanced models and maintain dialogue with outside researchers. Their letter presents their account of a dispute with the company. The circumstances of the terminations and the interpretation of internal procedures remain contested.