The letter: walked out by security
The three addressed their open letter to OpenAI's Safety and Security Committee, Safety Advisory Group and Mission Advisory Council, and published it on October 8. The account is vivid and uncomfortable.
Korbak describes being called into a meeting with the head of the safety department, told the company no longer trusted him, then escorted out by a security officer who took his badge. He learned Wang and Balesni had been fired the same way. "If conduct that was considered normal last month now constitutes grounds for sudden dismissal, everyone at OpenAI is left guessing where the line is," the letter reads.
Balesni, in a post accompanying the letter, was blunter: "I believe we were fired for prioritizing safety over the near-term interests of OpenAI as a corporation."
The response: a breach of trust, no receipt
OpenAI's statement, posted through @OpenAINewsroom on X and reported by Reuters on Friday, says a thorough investigation found the three violated "clear policies on handling sensitive information." The key line: "Our internal investigation uncovered a significant breach of trust beyond what's outlined in the letter they published and we stand by the decision to not continue their employment."
It also carried an explicit denial: "We have not and do not terminate any of our employees for raising concerns."
Read carefully, the two documents disagree about procedure more than about the underlying work. OpenAI asserts clear policies were violated. The researchers assert the rules were being written in real time around an unprecedented incident. Both can be sincerely stated. Only one can be the practice that was actually enforced, and there is no way to check without the specifics.
Two firings, two different stories
Korbak and Balesni were tied to the investigation of an incident involving OpenAI's agents and Hugging Face. Korbak says he served as OpenAI's primary technical contact for METR, the external safety lab that examined the incident. "To be clear, talking to METR was my job," he wrote on X. He says he was told verbally he was being fired over that communication, and nobody said exactly what he did wrong or wrote it down.
He also says he had raised concerns internally for months about OpenAI losing the ability to monitor what its agents "think" — chain-of-thought monitorability, one of the few reliable tools for catching AI systems behaving badly. He believes that was the real reason he was let go.
Wang's case, according to the researchers, is a separate matter entirely. She had delegated access to an executive's email inbox for recruiting purposes, and IT never removed it despite her requests. When she accidentally opened a sensitive email, she reported it within minutes. If that account holds, one of the three was fired for reporting a data-access lapse, not committing one.
The three demands
The letter asks OpenAI for three things: honor its public commitments to embed external safety auditors like METR with employee-level access inside the organization; preserve the monitorability of frontier models; and clearly define how employees are allowed to work with outside safety groups.
OpenAI, in its statement, confirmed it is working on contracts with external safety auditors and agreed that frontier model monitorability needs industry-wide commitment. That is a genuine concession sitting inside a rebuttal. It also sharpens the researchers' point: if the company agrees auditors need formal access, the question becomes what the rule was while that access did not exist.
What would settle this
Our coverage of this firing on October 3 ended on a single condition: say what was shared. Six days later, both sides have gone public in detail, and OpenAI still will not say what information left the building or what the "breach of trust" consisted of. The researchers have been specific about their own conduct. That asymmetry is now the story: one side describes everything except the allegation; the other describes the allegation and nothing else.
A practical note for everyone outside the building: the rules for how frontier labs share safety findings with outside evaluators are evidently still being written, sometimes after the fact. Anyone building on these models should expect the boundary to be blurry and keep the audit trail.
Sources
- [1] Reuters via BusinessWorld Online — OpenAI says it has fired three researchers for violating sensitive information policy (Oct 9, 2026)Read source
- [2] Reuters via Northland News Radio — OpenAI says it has fired three researchers for violating sensitive information policy (Oct 9, 2026)Read source
- [3] Reuters via The Hindu BusinessLine — OpenAI says it has fired three researchers for violating sensitive information policy (Oct 9, 2026)Read source
- [4] The Decoder — OpenAI's safety crisis keeps getting worse and the company keeps making it worse (Oct 9, 2026)Read source
- [5] RuntimeWire — OpenAI says fired safety researchers broke trust, denies retaliation (Oct 9, 2026)Read source
- [6] ODSC — OpenAI Denies Firing Researchers Over AI Safety Warnings (Oct 9, 2026)Read source
- [7] alfw.ai — OpenAI Fired the People It Hired to Keep It Honest (Oct 3, 2026)Read source