A diary, a threat, and a $30 billion stake: What the Musk vs OpenAI trial has actually shown so far
(Disclaimer: This is image is generated using AI)
Two days before opening arguments, Elon Musk pulled out his phone and threatened OpenAI’s president by text. Greg Brockman had floated the idea of a peace deal—both sides drop everything, walk away. Musk’s reply, now sitting in the trial record: “By the end of this week, you and Sam will be the most hated men in America. If you insist, so it will be.”The thing about the trial that began on April 27 in an Oakland federal courtroom is that it’s full of moments like this. Moments where you remember the people building the future of artificial intelligence are also, on a very human level, deeply not okay. They keep diaries where they write things like “Financially, what will take me to $1B?” They tear paintings off walls during equity negotiations. They rant about whether their rivals at competing AI labs are evil. They threaten each other by SMS. And then they put on suits and try, with varying degrees of success, to convince nine ordinary people in California that they are the reasonable one.On paper, this is a narrow case. Musk dropped his fraud claims days before the trial started. Two claims survive. He wants $150 billion in damages, Sam Altman removed from OpenAI’s board, and the for-profit structure the company adopted last year unwound. What’s actually playing out, in front of Judge Yvonne Gonzalez Rogers, is something looser and meaner. After roughly seven trial days, neither side looks like it’s running away with this. Both sides have witnesses who damaged themselves on the stand. The jury looks tired. And the strongest piece of evidence in the entire courtroom is a folder of text files on Greg Brockman’s laptop.
The man who isn’t testifying haunts every email
You can date the rupture between Musk and Altman, if you want, to May 2015. Altman emailed Musk pitching a “Manhattan Project for A.I.” Musk replied that evening. Probably worth a conversation, he said. By December that year OpenAI existed as a 501(c)(3) nonprofit. Musk had drafted most of the founding language, picked the name, and written to Altman that he’d dedicate “whatever amount of my time is useful,” even at the cost of SpaceX and Tesla.But the trial keeps revealing that the founding wasn’t really about charity. It wasn’t even, despite Brockman’s later courtroom telling, an Altman-and-Brockman buddy project that Musk happened to fund. It was a panic response. And the man who triggered the panic isn’t in the courtroom. He isn’t on the witness list. He runs Google DeepMind, and his name comes up more in Musk’s evidence file than almost anyone except Altman himself.Demis Hassabis founded DeepMind in 2010 and sold it to Google four years later. According to Brockman’s testimony, Musk talked about Hassabis “many, many times” during OpenAI’s early years and was “very consistent and fixated” on him. The first thing Brockman remembers Musk asking at one early AI dinner was, “Is Demis Hassabis evil?” In a 2016 email Musk wrote, “I feel like they are playing the Super Bowl and we are playing the Puppy Bowl.” Shivon Zilis—then an OpenAI board member, now the mother of four of Musk’s children—wrote him a personal note calling slowing Hassabis down “the only nonnegotiable net good action I can see.” She also passed along a rumour that DeepMind staffers had started meeting in a London coffee shop without their phones, because they didn’t trust Hassabis not to be reading their email. The last reference to Hassabis in the trial exhibits so far is a March 2019 message from Altman to Musk: “Have some mild Demis updates to share.” They agreed to discuss it on the phone.OpenAI was incorporated to stop Hassabis. The for-profit pivot, the Microsoft billions, the lawsuit, this courtroom—it’s all downstream of that one fixation. Once you see it that way, a lot of the testimony stops looking like a charity dispute and starts looking like a custody battle between two men who can’t agree on whose paranoia counts as the founding vision.
Musk on the stand, doing his worst impression of Musk
Musk was the first witness, on April 28. Black suit, black tie. By every account from inside the courtroom, he looked oddly flat. Elizabeth Lopatto of The Verge, who covered Musk’s defamation trial a few years ago, noted that he had charmed that earlier jury into acquittal. This time the charm did not arrive. He only really animated when bragging—about coming up with the name OpenAI, about recruiting Ilya Sutskever, about being instrumental.His direct examination ran the standard founder-myth playbook. He arrived in Canada with $2,500 in traveler’s checks. He works 80 to 100 hours a week. He doesn’t own a yacht. Neuralink’s actual long-term goal, he explained from the stand, is AI safety—if we can closely tie the human world to AI, “if there’s symbiosis we’re more likely to have a future with AI that’s good for humanity.” Sure.The cross-examination is what people will remember. OpenAI’s lead counsel William Savitt has, in a previous life, both represented Musk and sued him—Tesla’s side of the SolarCity case, then Twitter against Musk during the 2022 acquisition fight. He knows Musk’s tells, and he came in prepared. He pulled up a 2026 X post in which Musk wrote that Tesla “will be one of the companies to make AGI and probably the first to make it in humanoid/atom-shaping form,” and asked Musk to square it with his sworn testimony minutes earlier that Tesla wasn’t pursuing AGI. He pulled up the four-page term sheet for OpenAI’s 2018 for-profit pivot and got Musk, after several rounds of crosstalk, to admit he hadn’t actually read it. He took Musk’s deposition figure—$100 million in donations—and made him walk it back, on the stand, to $38 million.Musk handled none of this well. He compared one of Savitt’s lines to asking when he had stopped beating his wife. The judge cut him off. He raised his voice. He volunteered, completely unprompted, that his “AI-enabled robot army” remark about Tesla wasn’t meant in a military sense. The judge then asked him to summarise the plot of Terminator in one sentence. “Worst-case situation is AI kills us all, I suppose,” he said. The jurors had developed what one reporter called “thousand-yard stares.” Lopatto’s verdict, after watching five hours: “I have never been more sympathetic to Sam Altman in my life.”Then Musk’s fixer answered a question he was not supposed toThe strangest hour of the trial happened the next day. Musk had stepped down. Jared Birchall, who runs Musk’s family office, took the stand. Most of his testimony was so dry that a woman in the gallery pulled a sleep mask down over her eyes.Then, near the end of his direct examination, a note was passed to the lawyer questioning him. The lawyer asked Birchall about the $97.4 billion bid Musk and a coalition had made for OpenAI’s nonprofit assets in February 2025. Birchall started talking. He said the bid existed because Sam Altman was on “both sides of the table”—the for-profit and the nonprofit—trying to undervalue the charity’s stake during the company’s restructuring.OpenAI’s lawyers leapt up. The jury was sent home early. Then Judge Gonzalez Rogers did something unusual: she started questioning Birchall herself. Where did the $97.4 billion number come from? He didn’t know. Who chose it? The legal team. Who on the legal team? He couldn’t recall. Discovery on the bid had been blocked before trial. Now the question was whether Musk’s own lawyers had just pried that block open by asking about it on the stand.When the judge demanded to know who’d written the note that triggered the question, the courtroom went silent. Eventually Marc Toberoff, one of Musk’s senior attorneys, stood up and admitted it was him. Why? “I thought it was appropriate.”“Sounds like you wanted to open the door, then,” the judge said.Birchall’s testimony about the bid was struck. Whether more comes through that door is one of the trial’s open questions. So is the simpler one of why a billionaire’s most senior fixer claimed under oath he could not remember who picked a $97.4 billion number.
Brockman’s diary, and the cleverest boy in the room
Brockman took the stand in week two, in an unusual order—cross first, direct second. The cross was where the journal entries came out, and where Brockman, slowly and entirely on his own, made himself untrustworthy to the jury.His stake in the OpenAI for-profit is now worth roughly $30 billion. Steven Molo, Musk’s lead lawyer, asked Brockman why, if he himself had once written that $1 billion would be enough, he hadn’t donated the other $29 billion to the nonprofit. Brockman could have said the obvious things. That paper wealth isn’t liquid. That dumping that much stock at once would crater the company’s valuation. That founders are expected to keep skin in the game. He said none of that. He talked instead about how much the nonprofit’s stake in the for-profit was worth. Molo asked again. They went back and forth, and the jury’s heads turned side to side, and the question never got answered.Then there was the pedantry. When Molo read evidence aloud and skipped a small word—an “a,” a “the”—Brockman corrected him, every time. When asked whether Microsoft’s $10 billion investment was OpenAI’s biggest financial event, he replied that it was the only $10 billion investment. He kept saying things like “I wouldn’t characterise it that way.” Lopatto’s note: if you can define the word epistemology, you should not testify in your own defence.The journal entries did the rest. “It’d be wrong to steal the non-profit from him” sits uncomfortably close to Musk’s “stole a charity” framing. “Maybe we should just flip to a for-profit. making money for us sounds great and all” sits uncomfortably close to a confession. Brockman explained these were expressions of frustration, not plans. By the time he got to direct examination—where he spun a polished founding myth involving cozy dinners in Napa, traffic jams nobody noticed because the conversation was so good, and a photograph from OpenAI’s first day of work that very pointedly did not include Elon Musk—the jury had already heard him say, on cross, that he didn’t think it was morally bankrupt to promise Marissa Mayer in 2015 that he was personally donating $100,000 to the new lab and then never make the donation.His direct testimony did include one genuinely vivid moment. Recalling a 2017 meeting where the cofounders refused Musk’s demand for majority equity, Brockman said Musk tore a Tesla Model 3 painting off the wall and started storming out. “I actually thought he was going to hit me,” Brockman told the jury. As he was leaving, Musk asked Brockman and Sutskever when they were planning to quit. Then he said he’d withhold funding until they decided. The quarterly donations stopped after that.
What the jury actually has to decide, and why nobody looks good doing it
The witnesses still to come are stacked. Shivon Zilis is up first, with the possibility her testimony won’t be live-streamed because of threats made against her and her children. Microsoft CEO Satya Nadella is scheduled for then. After him, Ilya Sutskever, who co-founded OpenAI, was central to the November 2023 board move that briefly fired Altman, and according to filed evidence had vested OpenAI shares worth roughly $4 billion at the time of that firing. Why he flipped and pushed for Altman’s return is one of the most-asked questions in the AI industry. He may answer some of it under oath.The trial is expected to wrap by May 21. The jury will then be asked something fairly narrow—did OpenAI breach a charitable trust or a contract Musk relied on when he gave the company $38 million? If the jury finds for Musk, Judge Gonzalez Rogers decides damages and remedies, which on Musk’s wishlist include unwinding OpenAI’s for-profit structure entirely.Two facts hover over what’s actually at stake. SpaceX, which now owns xAI, has filed for an IPO that could reportedly raise up to $1.75 trillion. OpenAI is preparing one of its own. A verdict either way will land in the middle of that planning, and is likely to move markets faster than the appeals that follow.What the trial has actually shown so far is something the legal filings can’t quite capture. Both Musk and Brockman damaged their credibility on the stand, in different ways. Musk filibustered yes-or-no questions and forgot details he’d testified to hours earlier the same morning. Brockman argued over articles, refused to give a clean answer about his own $30 billion stake, and was caught wishing in his own diary for a way to “get out from Elon.” The jury, as Judge Gonzalez Rogers told Musk’s lawyers when they tried to remove jurors who disliked Musk, doesn’t have to like him. It just has to decide who is more credible.On the available evidence, that’s a hard call. Not because either man comes off well, but because neither does. Three weeks of testimony are still ahead. The strongest witness either side has put up so far is, of all things, a text file on Greg Brockman’s laptop, written nine years ago, by a man who probably never imagined a stranger would one day read it out in court.
