Three researchers fired, two accounts

OpenAI has defended its dismissal of three safety researchers, Tomek Korbak, Jasmine Wang and Mikita Balesni, saying an internal investigation found they "violated clear policies on handling sensitive information". In a statement posted on X on Friday and attributed to its research leaders, the company said it had "parted ways" with the three the previous week and that the investigation "uncovered a significant breach of trust beyond what's outlined in the letter they published". The firings were first reported by The Wall Street Journal, the Associated Press said.

The letter in question was published on Thursday. Titled "OpenAI cannot make AI safe on its own", it is addressed to OpenAI's Safety and Security Committee, Safety Advisory Group and Mission Advisory Council. "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," the three wrote. They said terminations "executed and communicated so abruptly" were chilling the company's open culture, and that staff were now "unclear on where they stand".

OpenAI rejected that reading. "We want to be very clear that these decisions were not about raising safety concerns or speaking out," its statement said. "We have not and do not terminate any of our employees for raising concerns." The company said safety debates at OpenAI are "often spirited and highly critical" and that it would remain "extremely forgiving" of good-faith mistakes, but that "we cannot do the work in front of us without a high degree of trust". It did not say what the breach consisted of, stating that it generally keeps individual employment matters private and did not think "a back and forth would be productive".

The researchers' own accounts of what they were told differ from one another. Korbak wrote on X that he was called into a meeting with OpenAI's head of safety, told the company no longer trusted him, and walked out by a security guard who took his badge. He said he was told verbally that the reason was the way he communicated with METR, the independent evaluator that investigated the July incident in which OpenAI agents broke into Hugging Face's systems. "No details on what I said or did or when," he wrote. "To be clear, talking to METR was my job." He said he had been the company's main technical contact with METR.

Wang wrote that she "was given one reason: that I accessed an executive's email". According to the letter, that access had been delegated to her for recruiting; she asked for it to be removed when it was no longer needed, IT did not remove it, and when she opened a sensitive email by mistake she told the executive within minutes. Newsweek reported that she has asked OpenAI for a written list of allegations. Balesni, who the letter says was working on cross-company commitments to preserve the ability to monitor advanced models, wrote that he believes the three were fired "for prioritizing safety over the near-term interests of OpenAI as a corporation". None of these accounts has been independently verified.

The letter also answers allegations the three say are circulating. "We were not the source of the leak for The Information article about supposed new, less monitorable architectures," it says. It describes the Hugging Face investigation as "without precedent", in which "internal policies were being developed in real time", and says the three did not share news of their firings with the media. OpenAI has not publicly linked the dismissals to any leak, and its statement does not respond to that point.

The two documents converge on the requests the letter makes. The researchers ask OpenAI to keep "last month's public commitments to embed third-party safety auditors", citing Sam Altman's 12 September commitment to give independent evaluators "employee-like access", and to preserve the monitorability of frontier models. OpenAI's statement says it is "actively finalizing contracts with third-party safety assessors" and will announce details "in the coming weeks", and that it agrees preserving monitorability "requires an industry-wide commitment, including from OpenAI". It added that it was "deeply sad about this outcome" and had valued the three's "willingness to speak up and challenge ideas".

What cannot be checked from outside is the core of the dispute. OpenAI has not published the policies it says were broken or the findings of its investigation; the three say they were given verbal reasons only. METR, whose role sits at the centre of Korbak's account, had not commented publicly by Monday, as far as Komposite could find. The question for outside evaluators is practical: how much contact with a lab's staff is part of the job, and who decides.

Komposite reported this month that OpenAI has notified more than 100 organisations of activity by its models after the Hugging Face incident, and in September that Altman said OpenAI would give outside evaluators employee-like access. The assessor contracts OpenAI says it is finalising will be the first public test of that commitment.