OpenAI research leaders answer the three fired safety researchers

On 9 October 2026, OpenAI’s research leaders published a note answering three safety researchers the company had fired the week before. The note, posted by OpenAI Newsroom at 06:17 UTC, names Jasmine, Mikita, and Tomek. Their letter, OpenAI cannot make AI safe on its own, is signed Tomek Korbak, Jasmine Wang, and Mikita Balesni and addressed to OpenAI’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council. The Verge reported the letter on 8 October 2026.

Neither side’s factual account of the dismissals has been independently verified. OpenAI says it generally keeps individual employment matters private, and that it does not believe “a back and forth would be productive or lead to a resolution.”

What OpenAI says

The note says OpenAI “parted ways” with the three “after a thorough investigation found they violated clear policies on handling sensitive information.” It says the investigation “uncovered a significant breach of trust beyond what’s outlined in the letter they published,” and that the company stands by ending their employment. Those sentences are OpenAI’s claims. The note does not name the policies or describe the conduct.

On the reason for the firings, the note says: “We want to be very clear that these decisions were not about raising safety concerns or speaking out.” It says safety debates at OpenAI are “often spirited and highly critical,” and that “We have not and do not terminate any of our employees for raising concerns.” The close repeats that the decisions “were not about them raising safety concerns.” OpenAI says it is “deeply sad about this outcome,” that it appreciated the three researchers’ “willingness to speak up and challenge ideas,” and that it had “placed enormous trust in them.” The note says the work cannot be done “without a high degree of trust,” and that OpenAI “will continue to be extremely forgiving of our team making good-faith mistakes.”

What the researchers say

The letter says communications around the firing have made former colleagues “afraid to speak and operate in ways that, until last week, were an integral part of working at OpenAI.” They write that they could raise safety concerns and draw on independent safety organizations, and that they acted within the norms of the time.

Their account of the conduct: they were not the source of a leak to The Information about “supposed new, less monitorable architectures,” and they do not believe they engaged with external parties outside the mandates of their jobs. The letter says Tomek Korbak was the technical point of contact for METR in the Hugging Face incident investigation and “did his best to handle this with care,” and the researchers fear the firings “may be used to justify ending OpenAI’s work with METR.” OpenAI’s note does not mention METR. The letter says Mikita Balesni worked with board members and the C-suite on cross-company monitorability commitments and “took care to remove sensitive details” before sharing materials. The letter also says Jasmine Wang’s access to an executive’s email “was delegated for recruiting purposes, with permission,” that IT did not remove it after she asked, and that she reported an accidentally opened sensitive email within minutes. That is their account, not a finding.

The three requests, and OpenAI’s replies

The letter makes three requests. OpenAI’s note does not address the email, METR, or leak details.

First, the researchers say OpenAI must keep commitments to embed third-party safety auditors. A third-party safety assessor, as both documents use the idea, is an outside organization brought in to audit or assess models. The letter points to what it calls Sam Altman’s 12 September commitment to give independent evaluators ongoing, employee-like access. OpenAI says it is “actively finalizing contracts with third-party safety assessors and will announce details in the coming weeks.”

Second, the letter says OpenAI must preserve monitorability. Monitorability, in the letter, means monitoring a model’s chain of thought. The researchers write that the industry does not yet know how to safely deploy models it cannot monitor. They attribute to Jakub a line that chain of thought monitorability is “fragile and unfortunately trending in a negative direction.” OpenAI’s note says: “We agree with the letter that preserving the monitorability of frontier models requires an industry-wide commitment, including from OpenAI.” It points to its Monitoring Monitorability work, open-sourced evals, the GPT-6 Astra system card, and its Alignment blog.

Third, the letter asks OpenAI to let safety researchers raise concerns and work with outside organizations under written rules. OpenAI’s note has no separate reply to this request. Under the assessor point it says “Many of our researchers already work with 3p safety organizations productively,” and elsewhere that safety and research debates “happen every day at OpenAI.”

OpenAI asserts a breach of trust and a policy violation. The researchers assert that they stayed inside the norms of their jobs and that colleagues are now afraid to speak. The dated follow-up is the assessor announcement OpenAI says will come in the coming weeks.