On the Wednesday of the trial’s third week, before the jury was let in, an OpenAI lawyer walked up to the bench with a small object wrapped in white cloth. Judge Yvonne Gonzalez Rogers unwrapped it on the desk in front of her.It was a gold statue. About the size of a Little League trophy. The rear half of a donkey, two legs and a tail and a hindquarter, mounted on a small white base. The inscription read: “Never stop being a jackass for safety.”The judge looked at it. She set it down. “I don’t want it,” she said.The trophy belonged to Josh Achiam, OpenAI’s chief futurist, who joined the company as an intern in 2017. He testified later that morning about an all-hands meeting in early 2018, the same meeting where Elon Musk announced he was leaving OpenAI to chase AGI at Tesla. Achiam had spoken up to say racing for AGI sounded reckless. Musk, by Achiam’s account, had snapped and called him a jackass in front of fifty or sixty people. At the next all-hands, Dario Amodei and a colleague had presented Achiam with the trophy, hand-engraved.Musk, on the stand in week one of this trial, had testified under oath that he never called anyone at OpenAI a jackass. He might have used some “strong language,” he allowed. He might have said “don’t be a jackass.” But not the word as a label, not aimed at a person. OpenAI’s lawyers wanted the statue in evidence specifically because it bore on the question.The judge ruled the testimony could come in. The trophy could not. The jury never saw it. Photographs leaked out of the courtroom anyway.Part one of this dispatch covered the trial’s opening week, when Musk himself took the stand and produced more than 200 instances of “I don’t recall” under cross-examination. Weeks two and three were less about the principals and more about the people around them. Former CTOs. Former board members. A chief scientist who has not spoken to his CEO in over a year. The mother of Musk’s four youngest children. A Microsoft CEO who would clearly rather have been somewhere else.The strange thing about a trial built on betrayal is how often it produces a different kind of evidence than the lawyers ask for. The lawyers in this case have spent fifteen days reading texts and emails aloud. What kept emerging from those texts and emails was not betrayal so much as documentation. Almost nobody in this story seems to have ever sent a message they would later be glad to see projected onto a courtroom screen.The jury begins deliberations on Monday. The verdict is theirs. The record is already public.
The witness was Shivon Zilis. The evidence was her own emails.
Shivon Zilis had served on OpenAI’s board from 2020 to 2023 and advised the lab since 2016. She is the mother of four of Musk’s children, three of them by IVF after what she described under oath as a one-off romantic encounter. She had originally signed on as a co-plaintiff to this lawsuit. She quietly dropped out before trial.She took the stand in week two in a black cardigan and black jeans. Her voice on the stand was small. She testified that her job for Musk had been to find bottlenecks across his entire AI portfolio—Tesla, Neuralink, OpenAI—and solve them, 80 to 100 hours a week. She called this his “maniac mode.” The gallery laughed once and then stopped.Zilis had been the only person taking proper notes during the cofounder negotiations of 2017 and 2018. This made her emails the trial’s most-quoted exhibits. OpenAI’s lawyer Sarah Eddy walked her through them one at a time, at the patient, almost soothing pace of a teacher returning a graded paper.A 2017 brainstorming email listed structural options for OpenAI. One read: “switch to for profit in next couple of weeks (woah fast!).” The phrase, written eight years ago by Musk’s then-girlfriend, sat on the courtroom screen for a long time.Another email, to Musk’s money manager Jared Birchall, explained that OpenAI’s cofounders were holding the line on one demand: “ironclad agreement to not have Elon (or anyone) have absolute control of AGI.” A third, from August 2017, flagged something the cofounders themselves did not yet know—that Musk had silently halted his quarterly donations. Zilis sent her note two weeks before he eventually told them.Other documents were harder to characterise as routine. In one brainstorm, Zilis floated folding OpenAI entirely into Tesla. In another, she asked whether they could “find a way to get Demis.”Part one of this dispatch noted Musk’s long fixation on Demis Hassabis of DeepMind, the same man Brockman’s diary entries recorded Musk asking about, the first time the two ever spoke, as “Is Demis Hassabis evil?” Zilis’s email confirmed the fixation had run through her too. If Hassabis “hung around E,” she had written, “perhaps it would force him to think about humanity more.”The exhibit that produced the longest silence in the room came from February 2018, around the time Musk left OpenAI’s board. Zilis: “Do you prefer I stay close and friendly with OpenAI to keep info flowing or begin to disassociate? Trust game is about to get tricky.” Musk: “Close and friendly, but we are going to actively try to move three or four people from OpenAI to Tesla.”Zilis told the jury she had meant to write “trust framework,” not “trust game.” Eddy paused, looked at her, and moved to the next exhibit.The exchange that may stay with the jury came from February 2023, as word leaked that Musk was building a rival AI lab. Zilis had texted a friend stored in her phone under the name “Shahini Rubicon Fluffer”—a name that produced visible delight in the press row—about her impending resignation from the OpenAI board. “When the father of your babies starts a competitive effort and will recruit out of OpenAI there is nothing to be done.” Her friend suggested Musk put her on the board of the new thing. Zilis said she was bummed.By Thursday’s closing argument, Eddy needed only one line to summarise two days of testimony. “Even the mother of his children can’t back his story.” Zilis had, in a sense, written it for her.
Sam Altman survived four hours on the stand. The damage came from the witnesses who weren’t there.
Both sides spent weeks two and three watching their own people testify against them. For OpenAI, that was Zilis. For Musk, it was a longer parade, most of it on video, most of it about Sam Altman.Mira Murati did not appear in person. She was at the Met Gala the week her deposition played in court, which in context felt almost too on the nose. Murati was OpenAI’s chief technology officer from 2022 to 2024, briefly its interim CEO during the November 2023 firing weekend, and now runs Thinking Machines Lab, a one-year-old AI company valued at around $12 billion.In her deposition, Murati described an incident in which Altman had told her OpenAI’s legal team had cleared a new AI model from going through the company’s deployment safety board. She had called the head of legal directly to check. He had told her no such thing had happened. Asked whether Altman had been telling her the truth, Murati said, flatly: “No.”Her text exchanges with Altman from that same November 2023 weekend produced the most quoted message in tech industry history this side of “concepts of a plan.” Altman, scrambling about whether he’d be reinstated, asked her: “Can you indicate directionally good or bad?” Murati replied: “Directionally very bad. Sam this is very bad.”The phrase had its own subreddit within twenty-four hours.Helen Toner, the former board member who had voted to fire Altman, described in her deposition a “pattern of behaviour” related to his “honesty and candour.” Toner also delivered the trial’s quietest knife, this one aimed at Murati. After Altman’s firing, Toner said, Murati had been “strikingly unsupportive” and “remarkably passive,” refusing to acknowledge publicly that her own conversations with the board had helped trigger the removal. “She was waiting to see which way the wind would blow,” Toner said. “She didn’t realise that she was the wind.”Tasha McCauley, another former board member, used the phrase “toxic culture of lying.” Rosie Campbell, a former safety researcher, testified about her own concerns. Daniel Kokotajlo, another safety researcher, did the same by video.By Tuesday of the third week, the depositions had been hanging in the air for over a fortnight when Steven Molo finally got Altman himself on the stand.Molo opened the cross with a piece of theatre. He stood up, faced the jury, and read aloud a list of names.Mira Murati.Helen Toner.Tasha McCauley.Ilya Sutskever.Dario Amodei.Daniela Amodei.Elon Musk.Seven people who built OpenAI with Altman or worked closely under him. Seven people who have publicly described him as dishonest. Two of them now run Anthropic. One runs Thinking Machines. One runs Safe Superintelligence. The rest are no longer, in any meaningful way, part of his life.Altman, in a navy suit and a lavender tie, sat still. He told the jury he believed himself to be an honest businessman.The disclosures that followed were the more consequential part of the cross. Under his own lawyer’s questioning earlier the same morning, Altman had quietly walked back a 2023 statement to the U.S. Senate in which he had claimed he held no equity in OpenAI. He had “misspoken,” he explained. He did hold a passive stake via Y Combinator.He also conceded, in the same hours, a roughly $1.6 billion personal stake in Helion, a nuclear fusion startup that has since signed a deal with OpenAI. A $600 million stake in Stripe, which has a deal with OpenAI. A stake in Reddit, which signed a licensing agreement with OpenAI in 2024. A stake in Cerebras.Whether all of this matters legally is for the jury. What is no longer in dispute is the parallel pressure now building outside the courtroom. By Tuesday evening, the U.S. House Oversight Committee had opened a probe into Altman’s potential conflicts of interest ahead of OpenAI’s planned IPO. Six Republican state attorneys general had written to the SEC asking for scrutiny of the same conflicts. The Wall Street Journal published a detailed investigation by Wednesday morning.Musk’s strategy throughout this lawsuit has been openly two-pronged. He wanted a verdict. He also wanted Altman scrutinised. The first is undecided. The second, by any visible measure, is well underway.
Ilya Sutskever owns $7 billion of a company he says is run by a liar
Of all the witnesses across three weeks, Sutskever was the only one who looked, in the moment of his testimony, like a person you might trust to design a powerful technology.He took the stand on Monday of week three in a blue button-down shirt with no jacket. He was the only male witness in fifteen days of testimony to appear without one. He spoke softly. He made very little eye contact with anyone in the room. He looked unhappy throughout, in a way that was not theatrical.He has not spoken to Altman in over a year. He has not spoken to Brockman in over fifteen months. OpenAI’s lawyer told the judge he should be treated as a hostile witness.Two disclosures from his testimony will outlive the trial regardless of how the jury rules.The first is that his vested shares in OpenAI’s for-profit arm are worth, by his own disclosure under oath, about $7 billion. They were worth $5 billion in November. They have appreciated quietly while he has built a separate company, Safe Superintelligence, that has so far released no product but is valued at $32 billion. He is probably OpenAI’s third-largest individual shareholder, behind Altman’s indirect stake and Brockman’s $30 billion. He has not worked at the company for two years.The second is that Sutskever confirmed, in his own voice and under oath, every line of the fifty-two-page memo he had written for OpenAI’s board in 2023. The memo accused Altman of “a consistent pattern of lying” and of “undermining his execs and pitting his execs against one another.”He had spent roughly a year before the November 2023 firing collecting evidence on Altman’s conduct. He had not shown any of it to Altman directly. “Because I felt that, had he become aware of these discussions,” he said on the stand, “he would just find a way to make them disappear.”His justification for then flipping and supporting Altman’s return was that he had panicked. “I felt that, had I not done this, the company would have been destroyed. I felt that this was a Hail Mary.”The most striking moment of his testimony came almost in passing. Judge Gonzalez Rogers, who has spent three weeks resisting any drift into AI futurism, asked Sutskever an unusually open question. How would he compare the AI of 2018, when Musk left OpenAI, to the AI of today.Sutskever paused. He looked at her. “I would describe it as the difference between an ant and a cat.”OpenAI’s lawyers were paying attention. In their closing arguments three days later, they put up an image of a tiny ant beside a large pink cat on the courtroom screen. The jury laughed.Sutskever’s testimony left a question open that the trial does not need to answer. If the chief scientist of OpenAI, the man who recruited most of its researchers, the man who turned down a $6 million salary at Google to join it in 2015, no longer trusts the company’s CEO—but is also about to collect $7 billion from the company’s continued success—what does that say about the company? What does it say about him?The trial has not resolved it. It may not be the trial’s job to.
Satya Nadella came to court hoping nobody would notice
Microsoft is technically a defendant in this trial, accused by Musk of having aided and abetted OpenAI’s alleged breach of charitable trust by pumping in more than $13 billion. Their strategy has been a corporate yoga pose. Present, but not engaged. Every cross-examination conducted by a Microsoft lawyer has followed roughly the same shape. Was Microsoft there? No. Was Satya Nadella there either? No. Next witness.Nadella himself took the stand on Monday, May 11, in a navy suit and a powder-blue tie, and stuck to the pose. He said he was “very proud” Microsoft had taken a chance on OpenAI when nobody else would. He said he had never received a complaint from Musk about Microsoft’s investments, despite the two having each other’s phone numbers.That last detail, delivered casually with a glance at the jury, was good enough that three different outlets ran it as a headline.His one moment of genuine annoyance came on the subject of the November 2023 firing. Asked why he had supported Altman’s reinstatement, Nadella said he had repeatedly asked the OpenAI board for specifics on what Altman had done wrong and had never been given any. “It was sort of amateur city, as far as I’m concerned.”Both sides took the line as helpful to their case. For Musk, it suggested OpenAI was institutionally unstable, propped up by a Microsoft CEO who had to step in. For OpenAI, it was a Fortune 500 chief calling the board that fired Altman amateurs, which by implication meant Altman had been wronged.A 2022 internal Microsoft email also surfaced. “I don’t want to be IBM and OpenAI to be Microsoft,” Nadella had written to his executives. He explained on the stand that he had meant Microsoft should retain “self-sufficiency”—that the partnership should not become a dependency. He also conceded under questioning that by 2023, Microsoft was projecting $92 billion in returns on a $13 billion investment.A calculated bet, as he called it. One that has paid back so far at roughly seven to one.
Closing arguments began with a stuck TV. They ended with the plaintiff in Beijing.
Thursday morning’s closing arguments began with a forty-minute argument about a television.Steven Molo, Musk’s lead counsel, had wheeled a 36-inch external monitor into the courtroom to support his presentation. He had not consulted OpenAI’s lawyers on its use, despite the judge’s standing instruction that the two sides coordinate on anything visible to the jury. The judge looked tired. “What do I always tell you when you come in here,” she said. “Talk to the other side.”OpenAI’s team said they didn’t have the cable to use the monitor themselves and weren’t particularly interested in trying. Fifteen lawyers stood in the middle of the courtroom for several minutes debating dongles. Eventually one of Musk’s own lawyers carried the TV out of the room upside down, the cable dragging on the carpet behind him.Sam Altman walked in just as the TV walked out.Molo’s closing ran roughly two hours. He referred to Brockman at one point as “Greg Altman.” He told the jury Musk was not asking for money, which prompted the judge to correct the record after the jury left for break—Musk is in fact asking for $134 billion in damages.He produced an extended metaphor about a wooden bridge across a hundred-foot gorge, with a sign at the entrance reading: “Don’t worry, this bridge is built on Sam Altman’s version of the truth.” Would the jury walk across it? “I don’t think many people would.”He attacked Altman’s credibility for ninety minutes. He called Brockman’s $30 billion stake “morally bankrupt.” He insisted the November 2023 firing was the moment OpenAI had truly betrayed its mission. He left it to the jury to decide whose version of events to believe.Sarah Eddy went next for OpenAI, at the same patient pace she had walked Zilis through her emails two weeks earlier. She showed Musk floating a for-profit conversion in 2017, demanding majority equity, and souring on the idea when he was not given control. She read out a 2018 email in which Musk wrote that he would no longer invest in startups he did not control. She used the “even the mother of his children can’t back his story” line.She finished in under fifty minutes.William Savitt followed. He noted that Musk had used the phrase “I don’t recall” between 150 and 200 times during his testimony in week one. He noted that Altman and Brockman were both present in the courtroom and Musk was not.”Mr Musk came to this court for exactly one witness,” Savitt said. “Himself. Now he’s in parts unknown.”Musk was, at that moment, in Beijing. He had flown out with Donald Trump on Air Force One during week three, in apparent defiance of an order from the judge that he remain on recall in case he was needed again. His own lawyer apologised for his absence that morning. The judge said nothing on the record.Microsoft’s lawyer got up last and made what amounted to a five-minute case that the company should not have been a defendant in the first place. He sat back down. The jury was sent home for the weekend.
The jury starts Monday. The record is already done.
Deliberations begin Monday. The verdict is advisory. Judge Gonzalez Rogers will decide the legal questions herself, and if Musk wins any of his claims, she will rule on remedies in a separate phase that begins the same day.Musk wants $134 billion redirected to OpenAI’s nonprofit, Altman and Brockman removed from their roles, and OpenAI’s for-profit structure unwound. OpenAI wants to keep all three, and to proceed toward an IPO that could value the company at over a trillion dollars by year’s end.The legal questions are genuinely open. The statute of limitations is contested. The contract question is harder still. There is no written founding agreement that says OpenAI must remain a nonprofit. Musk’s case rests on emails, blog posts, and the meanings of conversations that happened a decade ago between four men who can’t agree on what was said. OpenAI’s defence is that Musk wanted the same things he is now suing them for doing, but wasn’t given them. Both can be true. The jury has to pick a frame.What weeks two and three actually produced, regardless of which frame gets picked, is a record.Seven former colleagues went on the record describing Altman as dishonest. A House Oversight inquiry into his personal investments opened mid-trial. Six state attorneys general wrote to the SEC. A former chief scientist confirmed under oath the contents of a 52-page memo about him, while also disclosing a $7 billion shareholding in the company that memo had been meant to reorient.On the other side, Musk’s longest-running confidante was shown, on documents she herself authored, to have functioned as a sustained channel of information back to him during her three years on OpenAI’s board. His testimony that he never called anyone a jackass now sits next to a small gold trophy currently in the possession of OpenAI’s lawyers. His own emails were used to argue he wanted the exact for-profit restructuring he is now suing to undo. His absence from closing arguments will be the last thing the jury sees of him.The trial began as a fight over the future of a charity. What it has produced, three weeks in, is a much smaller and more familiar document. A record of how the people running the most powerful technology of our time tend to behave when they think nobody is watching. They keep diaries. They text each other “directionally very bad.” They commission small gold trophies of donkeys’ backsides. They negotiate equity structures in haunted mansions over whiskey. They threaten each other by SMS forty-eight hours before opening arguments.Part one of this dispatch ended with the observation that whoever wins, the AI industry comes out of this trial smaller than it went in. Three weeks later, that still feels like the most honest summary on offer.The donkey trophy is back in OpenAI’s lawyer’s office. The judge has confirmed, on the record, that she does not want it. Whichever way the jury rules on Monday, somebody is going to have to find somewhere to put it.