Week 1 went badly for Musk. Week 2 went badly for Sam Altman.
As we covered Monday, the opening of Musk v. Altman featured Musk's own tweets and unread documents being wielded against him. But testimony this week shifted the spotlight, and what came out isn't something OpenAI can easily explain away.
Former CTO Mira Murati was the key witness. In a videotaped deposition presented Wednesday, she was asked whether Altman had told her the truth when he claimed OpenAI's legal team determined a new AI model didn't require review by the company's Deployment Safety Board.
Her answer: "No."
That board exists to catch real problems before models reach the public. OpenAI's own safety evaluations for GPT-5.5 show why it matters: the model scored "High capability" in both biological and cybersecurity threats, and even with the full review process, the UK's AI Safety Institute still found a universal bypass across every malicious cyber query they tested. Skipping that kind of review isn't a paperwork shortcut.
Murati described a broader pattern. "My concern was about Sam saying one thing to one person and completely the opposite to another person," she testified, calling it "a very difficult and chaotic environment to work with."
What gives her testimony weight is what she did next: she supported Altman's return anyway. After the board fired him in November 2023, Murati told the court the company was at risk of collapsing entirely. In text messages to Microsoft CEO Satya Nadella during the crisis, she stressed how important it was not to lose researchers to competitors. She backed Altman not out of trust, but because letting the company implode was the only worse option.
Murati now runs Thinking Machines Lab, a competitor that closed a $2 billion seed round at a $12 billion valuation last year. She's not a neutral party. But her credibility in court comes from exactly that complexity: she caught Altman in a lie and worked with him anyway because the stakes were too high not to.
Then came Greg Brockman's diary entries. OpenAI's co-founder and president was grilled over private journal entries from 2017, written while he was accepting charitable donations from Musk to fund what was supposed to be a nonprofit:
- "Financially, what will take me to $1B?" From a co-founder of a nonprofit.
- "It would be wrong to steal the non-profit from Musk. To convert to a b-corp without him. That'd be pretty morally bankrupt." He knew what they were doing.
- "Cannot say that we are committed to the non-profit. Don't want to say that we're committed. If three months later we're doing b-corp then it was a lie." He was planning for it anyway.
Brockman has since accumulated roughly $30 billion in net worth from his OpenAI stake without personally investing any capital. He also held a roughly $10 million stake in Altman's family office, a financial tie Musk's legal team used to argue his loyalties were compromised from the start.
Former board member Helen Toner added another thread, testifying that she "was used to the board not being informed about things." The same lack of transparency, described by different people, years apart.
The financial stakes make all of this more than courtroom drama. Musk donated between $38 million and $44 million to OpenAI between 2016 and 2020. The company is now valued at roughly $852 billion, and its most recent funding round included $50 billion from Amazon, with $35 billion contingent on OpenAI going public or achieving AGI. Musk is seeking up to $150 billion in damages.
This trial is airing out something neither camp wants on the record: the moment OpenAI stopped being a nonprofit in spirit, years before the paperwork caught up. Brockman was journaling about hitting a billion dollars in 2017. Altman was allegedly sidestepping safety reviews. And the most credible witness so far backed Altman's return not out of trust, but because the alternative was watching the whole thing collapse. The question for the jury isn't really whether Musk got cheated out of $44 million. It's whether anyone at OpenAI ever actually put the mission first.
