OpenAI fires three safety researchers who dispute misconduct claims and warn of chilling effect
Three former OpenAI employees say they were terminated for prioritizing safety over corporate interests, while the company maintains they were fired for mishandling sensitive information.
Three former OpenAI employees have published an open letter stating they were fired for prioritizing safety over near-term corporate interests, directly disputing the company’s claim that their dismissal was for policy violations. Mikita Balesni, Tomek Korbak, and Jasmine Wang — identified in Al Jazeera’s report on their letter — say their termination came after they worked with external safety evaluators, a practice they describe as previously encouraged at the company.
OpenAI’s position, detailed in the same Al Jazeera coverage, is that the three researchers were fired for violating policies on handling sensitive information and committing what the company termed a significant breach of trust. The company explicitly denies that safety concerns played any role in the decision.
The researchers’ letter, as reported by TechCrunch, rejects this framing entirely. They contend that behavior that was normal a month ago has suddenly become grounds for dismissal, leaving employees unclear on where they stand. The letter warns that this signals a chilling effect on OpenAI’s culture — specifically around AI safety work that might conflict with commercial priorities.
What makes this dispute particularly sharp is the researchers’ claim that they were doing exactly what the company had previously encouraged. Working with external evaluators, they argue, was standard practice until it wasn’t. That shift — from encouraged to fireable, with no clear boundary announced — is the core of their complaint about cultural chill.
The case sits at an uncomfortable intersection. AI safety research often requires external scrutiny and independent evaluation to be meaningful; a company simultaneously pursuing rapid product development and commercialization has obvious incentives to control the timing and framing of such scrutiny. The researchers are essentially claiming that OpenAI’s incentives have started winning, and that safety work is being redefined as misconduct when it proves inconvenient.
Whether their firing was genuinely about information handling or about the content of their safety work is probably unprovable from outside. What the letter makes clear is that the researchers believe it’s the latter, and that they see their case as part of a pattern where safety culture at OpenAI is being systematically eroded. For a company whose public positioning has long emphasized its commitment to safe AGI development, that’s a damaging accusation — and one that will be read carefully by the safety researchers still inside.