The courtroom receipts say what the November 2023 press releases couldn't: OpenAI's board fired Sam Altman because he lied, and his own CTO just said so under oath.
The Summary
- Mira Murati testified that Sam Altman lied to her about safety approval processes, falsely claiming legal had cleared a model to skip the deployment safety board
- Text messages from the 72-hour CEO shuffle show Murati dismissed interim CEO Emmett Shear as "rando Twitch guy", revealing the chaos inside OpenAI's leadership during the coup
- The depositions in Musk v. Altman are providing the first concrete evidence behind the board's vague "not consistently candid" firing rationale from November 2023
- What looked like a power struggle is turning out to be a pattern: safety guardrails bypassed, internal processes ignored, trust eroded at the exact moment OpenAI is racing toward AGI
The Signal
Mira Murati's deposition cuts through eighteen months of speculation. When asked directly if Sam Altman told her the truth about a new AI model's safety clearance, she said no. The model in question hadn't gone through OpenAI's deployment safety board. Altman told her it had legal approval to skip that step. It didn't.
This isn't about workplace drama. This is about the CEO of the world's most influential AI company allegedly lying to his chief technology officer about whether safety protocols were followed before releasing new capabilities into the wild.
"The board's strikingly vague explanation finally has specifics attached to it."
Murati also testified that Altman made her work more difficult during her tenure, though she qualified her criticism as "completely management related." Translation: not about technical vision or AI safety philosophy, but about how decisions actually got made inside the building. When your CTO says the CEO makes her job harder, that's a red flag. When she says it under oath in a lawsuit, it's evidence.
The text messages tell a parallel story about governance collapse. When the board appointed Emmett Shear as interim CEO, Murati immediately texted Altman: "New guy is rando twitch guy." Shear, for context, had just spent eight years running one of the internet's largest live-streaming platforms and was a Y Combinator alum. Not exactly a random pick. But to Murati, anyone outside the OpenAI inner circle was a rando.
Key details from the 72-hour shuffle:
- Shear lasted three days as CEO before Altman's return
- The appointment happened without buy-in from senior leadership
- Internal communication was so fractured that the CTO was learning about CEO changes in real-time and mocking them via text
The Musk v. Altman trial is doing what no OpenAI blog post ever would: forcing the actual decision-making process into public view. November 2023 was the AI industry's biggest soap opera because it played out on Twitter while almost no one knew what was actually happening. The board said Altman wasn't "consistently candid." Altman's allies said it was a coup. Microsoft scrambled. Employees threatened to quit en masse.
Now we know the board had cause. Murati's testimony suggests Altman bypassed safety review processes and misrepresented approval chains. That's not a personality conflict. That's a pattern of behavior that directly contradicts OpenAI's stated mission of building safe AGI.
"What looked like chaos was actually OpenAI's immune system trying to work."
The irony: Barry Diller testified he trusts Sam Altman, but added that "trust is irrelevant" as AGI approaches. He's right. Individual trust doesn't scale. Institutions need processes that work even when trust fails. OpenAI's safety board was supposed to be that process. If the CEO can tell the CTO it's been cleared when it hasn't, the process is decorative.
The Implication
If you're building AI agents, deploying models, or making infrastructure bets on OpenAI, this testimony reframes the risk. The question isn't whether Sam Altman is a good CEO or a bad CEO. The question is whether OpenAI has functional institutional controls as it approaches systems that could reshape the economy.
Murati left OpenAI in September 2024. So did other senior safety researchers. The pattern suggests OpenAI's safety culture isn't just weak, it's been actively undermined. For developers building on GPT-5 or GPT-6, that means your model provider's internal checks might be theater. For enterprises evaluating OpenAI against Anthropic or other labs, governance structure just became a first-order question, not a nice-to-have.