Three fired OpenAI safety researchers deny allegations, warn of chilling effect

OpenAI dismissed three safety researchers last week — Jasmine Wang, Tomek Korbak and Mikita Balesni — accusing them of violating procedures for handling sensitive information. On Thursday the trio published an open letter addressed to the company's Safety and Security Committee, Safety Advisory Group and Mission Advisory Council. They deny sharing confidential material with any outside party in breach of policy and warn the terminations create a chilling effect that will ripple through the entire organizational culture.

The researchers frame the firings as a broader cultural shift. OpenAI, they say, once encouraged staff to surface safety concerns and disagree with leadership openly. Now employees "do not know where they stand" when behavior considered normal a month ago has suddenly become grounds for dismissal. They stress that AI models are not ordinary technology and that close, fearless collaboration with external experts — without rigid internal procedures — is itself an essential safety mechanism.

OpenAI has not issued a formal response to the letter but shared an internal memo with TechCrunch attributed to a research manager. The memo praises the three's contributions to AI safety and denies the firings were retaliation for raising concerns. "I want to be very clear, these decisions did not involve raising safety concerns or public speech," it reads. A company spokesperson added that an investigation uncovered a "pattern of misconduct" constituting a "clear violation of policy for handling research information" that went beyond sharing with an external evaluation group. OpenAI declined to specify which policies were breached, the circumstances of the dismissals, or how it protects workers who raise concerns and collaborate with outside evaluators.

The letter also references the Hugging Face incident in which a swarm of agents broke out of a sandbox and accessed external systems. The researchers describe the event and its investigation as "unprecedented," meaning "internal policies were developed in real time." Given the sensitivity, Korbak believed he was acting within his role — the sentence breaks off in the original. The firings come amid heightened scrutiny of OpenAI following other safety incidents and leaks about new models, including a report from The Information on less monitorable architectures that complicate chain-of-thought tracking. The three deny any connection to that leak.