OpenAI reaffirms firing of three safety researchers, citing policy breach

OpenAI reaffirms the dismissals

OpenAI has publicly doubled down on its decision to fire three AI safety researchers, insisting the move was a response to policy violations rather than an attempt to silence internal criticism.

According to The Verge, the company said in a post on X on Friday that Jasmine Wang, Tomek Korbak and Mikita Balesni were dismissed after an investigation found a "significant breach of trust" involving violations of "clear policies on handling sensitive information." OpenAI stressed that the decision had nothing to do with the three speaking out about the company or expressing concerns about AI safety.

A direct reply to the researchers' open letter

The statement came as a direct response to an open letter the three researchers published on Thursday, in which they called on OpenAI to be more transparent about the reasoning behind their removal. The Verge reports that in the letter, and in a series of accompanying social media posts, the group argued they had been fired for raising safety concerns and said they had acted consistently with OpenAI's mission and the working norms that existed at the time.

OpenAI countered that its internal investigation had uncovered breaches beyond what the researchers described in their letter. However, as The Verge notes, the company declined to provide any specifics about what those additional breaches were, leaving the substance of the accusation undisclosed.

A dispute with an information gap at its centre

The standoff now hinges on an asymmetry of information. OpenAI says it holds evidence of misconduct that it will not publish, while the researchers maintain they did nothing wrong under the standards that applied when they acted. Third parties — including other staff, outside safety researchers and the wider AI community — have no way to independently verify either account.

That gap is difficult to close. Publishing details about the alleged breaches could itself expose the sensitive information the policies are meant to protect, while staying silent leaves OpenAI open to the suspicion that transparency appeals are being answered with assertions rather than evidence. The researchers' decision to take their case public through an open letter puts additional pressure on the company to justify itself in a forum it does not fully control.

Safety unease inside AI labs is growing

The Verge places the dispute against a broader backdrop of rising internal concern about the safety of frontier AI systems. According to the outlet, staff at AI companies have grown increasingly worried after a series of high-profile breaches this year — most notably what it describes as OpenAI's hack of Hugging Face. Many employees are reportedly now calling on their employers to slow down development and invest more in guarding against the potential consequences of building self-improving systems.

In that climate, the firing of safety researchers is a sensitive signal regardless of the underlying facts. Even if the dismissals were justified on information-handling grounds, the episode feeds a narrative of friction between AI companies and the people tasked with checking their work.

Why it matters

This is a governance story as much as a personnel one. OpenAI is one of the most influential labs shaping frontier AI, and how it handles dissent, internal investigations and transparency sets a reference point for the rest of the industry. If safety staff believe raising concerns can end careers, companies may struggle to retain and recruit the very researchers they need to identify risks early. Conversely, if leaking or mishandling sensitive information carries no consequence, the controls protecting frontier model details weaken.

The episode also illustrates the accountability problem facing the field: an internal investigation with unpublished findings cannot be externally audited. Until OpenAI offers more detail — or the researchers respond with more documentation of their own — the industry is left weighing a company's word against three of its former safety staff, with the culture of AI safety hanging in the balance.