Mira Murati testifies Sam Altman lied about AI model safety checks

It is one thing for tech executives to squabble in Slack messages, but it is an entirely different ballgame when a former CTO takes the stand and testifies under oath that her CEO lied to her.

In a video deposition played during the ongoing Musk v. Altman trial, former OpenAI CTO Mira Murati stated plainly that she could not trust Sam Altman. When asked under oath whether Altman was telling the truth regarding safety standards for a new AI model, Murati answered with a flat no.

A bypassed safety board

According to Murati, the confrontation centered around a new GPT model and OpenAI's deployment safety board. She testified that Altman falsely claimed the legal department had cleared the model to ship without needing to go through the safety board at all.

Murati smelled smoke and decided to check the story herself. She approached Jason Kwon, who was OpenAI's general counsel at the time and currently serves as chief strategy officer.

What she found was an immediate contradiction. Murati testified that Jason's account and Sam's account simply did not match. To protect the deployment, she stepped in and ensured the model went through the safety board anyway.

Murati stressed that her friction with Altman was entirely management-related. She explained that she was trying to do an incredibly hard job in a chaotic company, asking Altman to lead with clarity rather than actively undermining her.

The pattern of pitting execs against each other

Murati's testimony corroborates a long trail of internal alarms. Portions of a 52-page memo written by cofounder Ilya Sutskever were also read during depositions, accusing Altman of a consistent pattern of lying, undermining his lieutenants, and pitting his executives against one another.

Former board member Helen Toner shared a nearly identical story back in 2024, noting that executives had brought documentary evidence of manipulation to the board before Altman's brief ouster in late 2023.

Murati agreed with those descriptions on the stand. Interestingly, she also criticized the board's chaotic decision to fire Altman, noting that it put OpenAI at catastrophic risk of falling apart. Murati eventually left OpenAI in 2024 and launched her own venture, Thinking Machines Lab.

Why it matters

Under oath is very different from standard Silicon Valley spin, and this testimony strips away any remaining pretense about OpenAI's internal harmony.

For anyone building on top of frontier models, the details here are genuinely alarming. Safety boards exist to prevent hazardous releases. If the CEO is willing to falsely cite legal counsel to skip an internal review, it means the safety guardrails only held up because a technical executive took the initiative to double-check him.

We think this paints a troubling picture of Altman's managerial playbook. Pitting executives against each other and hoarding information might work when fighting hostile board battles, but it is a dangerous way to run an organization responsible for powerful AI systems.

It looks increasingly clear that the governance crisis at OpenAI was not just philosophical friction over non-profit ideals. It was a fundamental breakdown of trust between Altman and the people actually building the technology.