OpenAI is standing firm on its decision to fire three members of its safety ranks, escalating a rare public confrontation between the company and former employees who say they were pushed out for raising concerns about how it manages the risks of advanced artificial intelligence.
The dispute centers on Jasmine Wang, Tomek Korbak and Mikita Balesni, all of whom worked on alignment and safety questions at the ChatGPT maker. The company confirmed the dismissals in a post on X on Friday, saying an internal investigation had found the trio committed “a significant breach of trust” by violating “clear policies on handling sensitive information.”
OpenAI went further, preemptively rejecting the interpretation its former employees have advanced. The decision, the company said, “was not about raising safety concerns or speaking out.”
A fight playing out in public
The unusual exchange of statements began Thursday, when the three researchers published an open letter — along with a string of posts on social media — urging OpenAI to be more transparent about the terminations. They described the circumstances surrounding their departure as “suspicious,” and argued that the manner of their removal would have a chilling effect on employees who might otherwise flag safety problems internally.
“These decisions were not about raising safety concerns or speaking out.” — OpenAI
The company’s Friday post was written as a direct response to that letter. In it, OpenAI framed the firings as a straightforward matter of information handling rather than a referendum on internal dissent, insisting that its policies apply uniformly regardless of an employee’s views on AI risk.
The researchers, for their part, say the timing and the framing tell a different story. In their telling, the dismissals are the latest signal that safety-focused staff occupy an increasingly precarious position inside a company racing to ship frontier models.
How the story was framed — and why it matters
Coverage of the episode diverged sharply depending on the outlet, a split that reflects how contested the underlying facts remain.
- Corporate accountability framing: The Verge led with OpenAI “doubling down,” casting the X post as a defiant response to the researchers’ open letter and emphasizing the company’s insistence that the firings were unrelated to speech.
- Wire-service neutrality: Reuters reported the news in the plainest terms — that OpenAI said it had fired three researchers for violating its sensitive information policy — keeping the company’s account as the primary claim and treating the researchers’ rebuttal as secondary.
- Document-first framing: Several aggregators chose to publish the researchers’ letter in full, positioning the employees’ own words as the story and inviting readers to judge the company’s conduct for themselves.
- Chilling-effect framing: Tech-focused outlets foregrounded the allegation that the firings would suppress safety advocacy inside OpenAI, treating the episode as a case study in how frontier labs manage internal criticism.
That divergence is itself part of the story. With no independent account of the underlying investigation available, readers encounter the same three names and the same forty-eight-hour sequence of statements arranged to support very different conclusions about who is behaving badly.
The broader context
OpenAI has spent much of the past two years navigating departures, reorganizations and public criticism from people who once worked on its safety teams. Those exits have repeatedly become flashpoints in a wider debate over whether commercial pressure at leading AI labs is outpacing the guardrails meant to constrain them.
The company has consistently maintained that safety and product development are not in tension, and that it holds employees to the same confidentiality standards as any firm handling proprietary research. Sensitive-information policies at frontier labs typically cover unreleased model capabilities, internal evaluations and customer data — precisely the material whose disclosure could create competitive or security problems.
Critics counter that such policies are broad enough to be applied selectively, and that the burden of proof falls on the company when terminations coincide with internal safety advocacy.
What happens next
Neither OpenAI nor the three researchers have indicated that a resolution is imminent. The company has not published the findings of its investigation, and the former employees’ open letter asks for exactly that kind of disclosure — a request that OpenAI’s Friday statement did not address.
For the industry, the episode sharpens a question that has dogged every major lab: whether employees who believe a model poses a risk can raise that concern internally without jeopardizing their jobs. How OpenAI answers that question — in policy, in practice, and in public — will shape how the next generation of safety researchers decides where to work, and how loudly they are willing to speak once they get there.
