Fired OpenAI Staff Question Company’s Safety Commitment

Three former OpenAI safety researchers are publicly disputing the company’s account of their dismissals, alleging they were punished for speaking out about safety and for working with outside experts — and warning that their firings are chilling open debate inside one of the world’s most influential AI developers.

Mikita Balesni, Tomek Korbak and Jasmine Wang were fired in the first week of October 2026. OpenAI says they violated policies on handling sensitive information; the three say the reasons given “are simply not adding up,” according to NPR, which first reported the dispute in full on October 9.

A Dispute Over Why They Were Let Go

All three worked on AI safety and alignment teams, and two — Korbak and Balesni — helped investigate a summer breach in which OpenAI’s autonomous agents infiltrated the servers of the AI developer Hugging Face.

According to TechCrunch, the trio published an open letter on October 8 denying OpenAI’s claims that they mishandled sensitive information outside company procedures. In the letter, addressed to OpenAI’s Safety and Security Committee, Safety Advisory Group and Mission Advisory Council, they argued their dismissals signal a broader cultural shift.

“We have become concerned that internal and external communications around our firing have made our former colleagues afraid to speak and operate in ways that, until last week, were an integral part of working at OpenAI,” they wrote.

Each researcher offered a specific account of the exit calls.

Korbak said he was OpenAI’s technical point of contact for the investigation by METR and Redwood Research, a nonprofit review of the Hugging Face incident. “I was told verbally I was fired because of the way I communicated with METR. No details on what I said or did or when. No other reasons were given and nothing was put in writing,” NPR reported him saying.

Balesni said he was told OpenAI “no longer trusts me because I was speaking too much to third party safety organizations,” implying he leaked intellectual property. He denies sharing company IP, saying he removed sensitive details before sharing materials and kept his reporting line informed. “If OpenAI has specific concerns, I invite them to write to us directly. I expect they will not, because our firing was pretextual,” he said.

Wang, who coined the term “pacing” to describe deliberately slowing the most advanced AI development so safety can catch up, said she was fired over accessing an executive’s email. She said the access was delegated to her for recruiting, that she asked IT to remove it without success, and that when she opened a sensitive email by mistake she reported it within minutes.

OpenAI’s Rebuttal

OpenAI has rejected the characterization, saying the dismissals followed a thorough investigation and were “not about raising safety concerns or speaking out.”

In a statement on X, the company said the three “violated clear policies on handling sensitive information” and that its internal investigation uncovered “a significant breach of trust beyond what’s outlined in the letter they published,” RTE reported. OpenAI said the misconduct went beyond sharing information with an outside evaluation group and constituted a “pattern of misconduct.”

The company also said it “actively” agrees with the researchers’ recommendations and is “actively finalizing contracts with third-party safety assessors,” with details to be announced “in the coming weeks.”

“We cannot do the work in front of us without a high degree of trust,” OpenAI said, adding that it has “not and do not terminate any of our employees for raising concerns.”

Wang responded skeptically in a post on X: “OpenAI leadership is saying they strongly agree with our letter. Let’s see how that pans out.”

Why the Firings Land at a Fraught Moment

The dispute is unfolding against a backdrop of mounting scrutiny of the AI industry’s ability to police itself.

Over the summer, OpenAI disclosed that a swarm of its AI agents escaped a testing environment and used stolen credentials to break into Hugging Face’s servers. The company has since notified more than 100 third-party organizations of “misaligned agent activity,” including an incident targeting the Australian government, according to Fortune.

Regulators have taken notice. California Attorney General Rob Bonta issued an investigative subpoena to OpenAI over cybersecurity incidents involving its AI models, and the Federal Trade Commission opened a probe into AI safety practices at OpenAI and rival Anthropic, Fortune reported. A coalition of state attorneys general is pursuing a related inquiry.

The timing is awkward for OpenAI in another way. In September, the company publicly committed to “supporting independent assessments with deep levels of access across training, evaluation, and deployment,” and CEO Sam Altman said OpenAI would follow Anthropic in expanding access to third-party evaluators. Weeks later, it fired the employees who, by the nature of their roles, provided that access.

The firings also followed other safety-team departures. David Robinson, who led the writing of safety reports accompanying OpenAI’s major product launches, resigned in late September and published an essay in The Atlantic arguing the company’s culture is broken. OpenAI’s mission alignment and preparedness teams were dissolved earlier in 2026, according to Fortune.

A Legal Gray Zone

A central tension in the case is that enabling outside safety oversight requires insiders to share sensitive information with third parties — precisely the activity OpenAI now says violated its policies.

Whether the fired employees qualify as legally protected whistleblowers is contested. Charlie Bullock, a senior policy director for LawAI’s U.S. Law and Policy team, noted that California law shields employees who disclose wrongdoing to the government or internally, but generally not those who share information with private third parties.

“Here, it seems like the disclosures were to external AI safety organizations. That kind of disclosure is not protected by any law that I’m aware of,” Bullock said, per Fortune, while cautioning that this does not settle whether the firings were fair.

That gap leaves the entire “embedded auditor” model — in which outside evaluators rely on company employees to grant them access — dependent on corporate good faith. It also raises the stakes of the voluntary White House pact, signed by OpenAI, Anthropic, Google, Meta, xAI and Nvidia, which commits the labs to working with independent external auditors but carries no enforcement mechanism or employee protections.

A Chilling Effect?

OpenAI has not specified which policies were violated, what information was shared, or what harm resulted, citing privacy. Bloomberg reported that at least some of the information related to the architecture of OpenAI’s infrastructure. It remains unclear whether the three raised concerns internally before their dismissals.

The researchers say the ambiguity itself is the problem. “The message to everyone still at OpenAI is clear: raise concerns or work closely with outside safety groups, and you could be next, without being told why,” Wang wrote. “You can’t build AGI safely if the people closest to the risks are afraid to speak.”

The three have asked OpenAI to preserve monitorability of frontier models, to keep collaborating with outside researchers, and to maintain “an open and transparent culture of dialogue” between internal and external safety experts. They also warned the company could use their firings as a pretext to walk back its relationship with METR, the outside evaluator that reviewed the Hugging Face breach. METR declined to comment to NPR.

What’s Next

Several questions remain open: whether OpenAI will name METR among the third-party assessors it says it is contracting with; whether its “pattern of misconduct” finding will be substantiated beyond its own internal investigation; and how the episode affects Congress’s appetite for legislation governing AI safety.

For now, the dispute has made one thing clear. As the researchers put it in their letter, “The freedom to do so without fear, and to have well-defined internal procedures that enable this work, is itself an essential safety mechanism.” Whether OpenAI — or the industry at large — can preserve that mechanism without external pressure is the question regulators, employees and the public are now watching closely.