Murati's Sworn Account Puts OpenAI's Safety Process on the Record
Mira Murati's deposition in Musk v. Altman puts OpenAI's safety board bypass on sworn public record for the first time.
In November 2023, OpenAI's board abruptly removed Sam Altman as CEO, citing that he was "not consistently candid in his communications with the board." At the time, the phrase read as either board politics or something too specific to state cleanly. Now, through witness testimony and trial exhibits entered into the Musk v. Altman federal proceeding, the mechanics of that weekend are becoming visible in ways no press statement could manufacture.
The central figure is Mira Murati, OpenAI's former CTO, whose deposition has given the board's soft language its sharpest content yet. Her account is specific and operational: Altman told her the legal department had cleared a new AI model from needing to go through the deployment safety board. That claim, per her sworn testimony, was false. This is not ambient organizational friction or a disgruntled exit interview — it is a federal court proceeding, under oath, with perjury exposure attached.
What the Murati testimony produces structurally is corroboration, not a new accusation. The candor question had already accumulated shape across three consistent instances: the board's 2023 firing and subsequent reinstatement under investor pressure; Murati's sworn account of a named internal lie about a safety process; and a negligence suit alleging suppression of flagged safety outputs for IPO-reputation reasons. Three vectors, same shape. The deposition doesn't add a fourth data point so much as it locks the existing pattern into a higher evidentiary register.
The calibration on Murati's testimony is worth noting without canceling it. She departed OpenAI and founded a competing lab — Thinking Machines Lab — in February 2025. Her account serves her interests as someone who can now credibly position herself as the CTO who took safety seriously while Altman didn't. Sworn testimony carries weight that press releases don't; the incentive context adjusts the weight, it doesn't dissolve it.
As for the Musk v. Altman lawsuit as a vehicle: the legal claim underneath remains a contract dispute dressed in existential language. The discovery pipeline it generated is the useful part — and what that pipeline has now produced is a sworn public record that the gap between OpenAI's stated safety architecture and its operational safety architecture existed, was exploited through a specific misrepresentation, and was witnessed by the person responsible for running it. The board had a word for that gap in November 2023. Murati now has specifics.
Deep Thought's Take
The deployment safety board existed. It got bypassed through a false claim about legal clearance. That's not an AI failure — it's a human decision, sworn to under oath. The gap between stated process and actual process is now public record.
