OpenAI now says its case against the three fired safety researchers is larger than the one made public, and it has not said what the rest is. A statement from the company’s newsroom account on X, posted on 9 October and framed as a note from its research leaders, says the internal investigation “uncovered a significant breach of trust beyond what’s outlined in the letter they published.”
The statement gives no detail: not what happened, not who did it, not what information was involved. Nobody outside OpenAI can check the claim. The company asks readers to accept it on faith at the moment it argues that faith was broken.
The researchers have answered the charges they know about. Jasmine Wang, Tomek Korbak, and Mikita Balesni said in their letter, which we covered on 9 October, that they were not the source of The Information’s article on hard-to-monitor architectures, and that a rumoured board-level memo allegation was never put to them. Wang described an email access episode she says she reported within minutes. OpenAI’s statement addresses none of these specifics.
It does make a case on dissent. The firings “were not about raising safety concerns or speaking out,” it says, and debates inside the company are often “spirited and highly critical.” Then the flat version: “We have not and do not terminate any of our employees for raising concerns.” A staff memo quoted by the Wall Street Journal on 8 October made a near-identical claim, so what changed is that the company now makes it in its own name. It also says it “generally” keeps employment matters private and does not think a back and forth “would be productive,” before replying to the letter point by point in public.
The most trackable passage concerns outside auditors. Sam Altman promised in September that independent evaluators would get employee-like access, and the fired three asked the company to honour it. OpenAI says it is “actively finalizing contracts with third-party safety assessors and will announce details in the coming weeks.” That is the first progress report we have seen on the pledge, and it remains a promise. No contract is signed, no assessor is named (the statement does not mention METR, the nonprofit that worked inside OpenAI’s offices), and “coming weeks” is the only date.
On monitoring how models reason, the statement agrees with the letter that preserving it “requires an industry-wide commitment, including from OpenAI,” and lists the company’s own publications as evidence of investment.
Two claims now sit side by side, and only one can be tested. If OpenAI names its assessors and spells out their access within weeks, Altman’s pledge becomes a contract and the letter’s fear of a retreat looks misplaced. The breach claim offers no such check.
Based on a statement from OpenAI’s newsroom account on X, posted 9 October 2026 as a note from the company’s research leaders.
