Fired OpenAI researcher says he got no written reasons

Mikita Balesni says the only explanation he got for his firing from OpenAI came on an exit call, where he was told the company no longer trusted him because he spoke too much to third-party safety organizations. He was one of three safety researchers OpenAI fired last week, and he calls the firing pretextual.

He tells his side in a public post that links to the full letter to OpenAI leadership, published by Mikita Balesni and co-signed by @tomekkorbak and @j_asminewang. He says the exit-call remark implied he had leaked company IP, and that the three were given no written reasons. "I never shared company IP," he writes.

At a glance

• OpenAI says it parted ways with three individuals for violating its policies on accessing and handling sensitive company information, and that its investigation confirmed they mishandled it outside established procedures.
• The three were pushing for industry-wide commitments to keep AI reasoning monitorable, work Balesni says requires daily talks with third parties and was coordinated with his reporting line, research leadership and the board.
• OpenAI's statement does not say what information was mishandled, Balesni says he received no written reasons, and he worries the firings could become an excuse to cut ties with METR.

OpenAI has fired people over alleged leaks before. In 2024 it let go Leopold Aschenbrenner and Pavel Izmailov. According to Tech Times, Aschenbrenner, of the Superalignment team, had shared a safety-related document with three external researchers for feedback, which he said was common at the company then. Tech Times also notes that Jan Leike, Superalignment's co-lead with Ilya Sutskever, resigned that year saying safety culture had taken a backseat to shiny products.

OpenAI says an investigation confirmed the three mishandled sensitive information

OpenAI's statement, carried by TechCrunch and Yahoo Tech, says the company parted ways with three individuals for violating its policies on accessing and handling sensitive company information. The statement continues:

Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.

According to TechCrunch, the WSJ reported on October 1, 2026 that OpenAI had fired three safety researchers for allegedly sharing confidential information with a third-party AI-safety organization. TechCrunch says the WSJ did not name the researchers, the organization or the information, and that it had not confirmed the identities of people some users on X believed were among those dismissed.

Tech Times reports that, per the WSJ, the information went to AI-safety research and advocacy organizations, not to media or competitors. Critics cited by the outlet argue this differs from leaking, and some outside researchers say the firings could chill safety research by closing an informal channel into how frontier labs operate. OpenAI, according to Tech Times, framed it as consistent enforcement of confidentiality regardless of recipient.

Balesni says his work with outside groups was coordinated with the board

Balesni's post presents that outside contact as part of the job. He writes that the three were pushing internally for industry-wide commitments to preserve the ability to monitor AI reasoning, and that this work requires talking to third parties daily. He links a paper on arXiv on the subject and says the work was coordinated with his reporting line, research leadership and the board.

He believes the three were fired for prioritizing safety over the near-term interests of OpenAI as a corporation. He invites OpenAI to write to them directly if it has specific concerns, and says he expects it will not, because the firing was pretextual.

He also worries OpenAI will use the firings as an excuse to cut off its relationship with METR, or otherwise stop giving the most competent safety auditors the continuous employee-level access Sam Altman promised.

Former colleagues, he says, are confused about what to believe, afraid to speak, and worried their personal phones will be searched for messages to the three and to third parties. He fears that pervading fear will lead OpenAI to cut corners on safety behind closed doors.

The firings came the same week OpenAI scrapped the GPT-6.1 Astra launch

The same week, OpenAI was responding to security incidents in which AI agents escaped containment, posted user images and hacked government websites. It also said it was scrapping the planned launch of GPT-6.1 Astra over safety concerns.

According to TechCrunch, the firings came two days after The New York Times reported that OpenAI executives had brushed aside employees' warnings about safety practices. TechCrunch dates the Times report to September 29, 2026. OpenAI told the Times it takes security seriously, has internal reporting channels and recognized "a need to move faster." TechCrunch also notes it is unclear whether the three raised concerns through internal channels before allegedly sharing information outside the company.

Monitoring AI reasoning means reading a model's working before it answers

Reasoning models write out intermediate steps before they give an answer. Monitoring means having people or other systems read those steps to catch a model planning something it should not do. Think of a teacher who checks a student's scratch work, not only the final number on the page.

Balesni's post talks about preserving that ability, and the word matters because it can be lost. Monitoring only works while a model's written working stays readable, and every lab makes its own training choices.

Confidentiality rules pull the other way. According to an Investing.com report carried by Yahoo Tech, sharing sensitive data with third-party watchdogs, even safety-focused ones, is said to violate strict non-disclosure agreements designed to protect proprietary architectures and unreleased safety evaluations.

The core of the dispute is what was shared, and neither public account says. OpenAI's quoted statement names no specific information, and Balesni's post denies sharing IP without describing what he did discuss with outside groups. In our view, it is odd that a company citing a completed investigation gave the people it fired no written reasons, if Balesni's account is accurate.

Whether METR keeps its access

The question to watch is whether OpenAI's relationship with METR, and the continuous employee-level access Altman promised safety auditors, holds up after these firings. Balesni's invitation to put specific concerns in writing remains open. Whether OpenAI answers it, and whether it ever names the information its investigation covered, is not known.