Fired OpenAI Safety Researchers Dispute Misconduct Claims
Three recently terminated OpenAI safety researchers have issued an open letter disputing the company's misconduct allegations and warning that their abrupt firings will chill AI safety work.
Jasmine Wang, Tomek Korbak, and Mikita Balesni, three safety researchers dismissed by OpenAI last week, have published an open letter challenging the company's claims of misconduct. The trio rejected allegations that they mishandled sensitive data outside established procedures, warning instead that their abrupt terminations create what they described as a "chilling effect" on internal culture. OpenAI previously stated the researchers were let go following an investigation into a pattern of unauthorized information sharing.
In their letter, the researchers denied leaking details to the press regarding monitorability issues in OpenAI's latest models. They also clarified their actions during recent safety events. Korbak explained his external communications during a Hugging Face incident, where a swarm of agents breached their sandbox, were meant to build trust with outside evaluators. Balesni stated his work on AI monitorability was fully coordinated with OpenAI executives. Meanwhile, Wang clarified that her access to an executive's email inbox was originally authorized for recruiting and that she immediately reported accidentally opening a sensitive message after IT failed to revoke her access.
OpenAI has defended its decision, with a spokesperson pointing to a clear violation of policies regarding research information. An internal memo from a research leader asserted that the company does not terminate employees for raising safety concerns. However, the researchers argue that the sudden dismissals leave remaining staff confused about what collaboration with external safety groups is permitted, especially when such partnerships were previously standard practice.
For AI practitioners and safety auditors, this dispute highlights a growing friction between corporate intellectual property security and the collaborative nature of AI safety research. If researchers face termination for sharing data with external evaluators, it could severely limit third-party auditing and slow down the development of robust safety standards across the industry. The researchers urged OpenAI to honor its commitments to embed independent safety auditors and preserve the monitorability of its frontier models.
This is our own summary of reporting by TechCrunch AI
