Elon vs Sam, week 1: The first, most important AI trial of the decade. Funny enough it's not about AI

On Day 4 of the trial, Elon Musk (Tesla and SpaceX CEO, founder of xAI, OpenAI co-founder and early funder) admitted under oath that his AI company xAI is "partly" trained on OpenAI's models, and reporters in the room said the courtroom audibly gasped.

Pretty much every major AI lab does the same thing without making a fuss about it, including OpenAI itself, Anthropic, and Google. Legally the admission barely matters. In the room, though, it handed OpenAI's defence team a frame they could not have written better themselves.

This is the man who flew into a federal court in Oakland to call OpenAI a charity thief. He is suing for $134 billion in damages, wants Sam Altman (OpenAI CEO and co-founder) and Greg Brockman (OpenAI president and co-founder) removed, and wants the for-profit conversion unwound. On the witness stand, in the same week, his own attorney's setup question handed OpenAI's defence a frame that did not exist before, casting Musk not as a clean charity defender but as a charity beneficiary in the same breath.

The gasp itself was theatre, but the frame it created has stuck, and that frame is the strongest indication so far that the trial OpenAI is fighting is not the trial Musk thought he was filing.

A quick note on what this article is. I am not a lawyer, and I have no special expertise in charitable trust law. I have spent the last week trying to translate the legal language coming out of the Oakland courtroom into something I can actually understand, because this trial is going to shape how AI companies are built for the next decade. What follows is my attempt to lay out the facts in plain English, including a few legal terms I've had to look up myself.

Case 4:24-cv-04722-YGR opened on 27 April 2026 in front of Judge Yvonne Gonzalez Rogers (a federal judge in the Northern District of California, appointed by President Obama in 2011, best known for presiding over the Epic Games versus Apple antitrust case where Fortnite's maker forced Apple to open up its App Store payments). It is being tried as an "equity trial", which is a kind of court case where the judge has the final say rather than the jury. The nine people in the jury box are an advisory jury, which means they hear the evidence and make a recommendation, but Judge Gonzalez Rogers is not bound by what they decide, and any damages or remedies are hers to set alone.

Two claims survived to trial. The first is breach of charitable trust, which I had to look up myself. In plain English, when a nonprofit takes donations on the basis of a stated mission, the law treats that mission as a binding promise the entity owes back to its donors and to the public. Musk's case is that OpenAI broke that promise. The second is unjust enrichment, which is closer to common sense. It is the legal idea that you cannot keep financial gain you obtained unfairly at someone else's expense. Musk argues that Sam Altman, Greg Brockman, and OpenAI's investors got rich off a charity he funded, and that the law should claw some of those gains back. The original complaint had 26 separate legal claims, including racketeering and fraud, and most were dismissed before the trial even began. The breach of contract claim was thrown out in March 2025 because Musk could not point to a written agreement with OpenAI. He admitted on the stand he did not have one.

To be super clear, if Musk wins on either remaining claim, the proceeds do not go to him. They go to the OpenAI Foundation, which is the nonprofit entity that today still sits at the top of OpenAI's structure. The foundation already owns roughly 26% of the for-profit OpenAI Group PBC, a stake worth about $130 billion at current valuations, and its money is meant to be spent on broadly beneficial AI research, AI safety, AI literacy programs, and grants of that kind, in the same way the Bill and Melinda Gates Foundation spends down its endowment on global health. Musk has told the court in writing that he will not personally take a single dollar out of any win, and any damages, equity, or settlement that comes from the case flows back to the foundation rather than to him. That is partly tactical (it makes him look like he's fighting on principle, not for the cheque) and partly the way charitable trust law actually works, in that money recovered from a breached charity has to go back into the charitable mission, not into the donor's pocket. The remedies he is seeking are structural, namely the removal of Altman and Brockman, unwinding of the for-profit conversion, and return of any "ill-gotten gains" to the foundation.

The headline framing is a decade-old grudge match. The trap is mistaking that headline for the actual story, because the harder reading sits underneath it.

The strongest argument for converting OpenAI from a nonprofit into a for-profit was made by Elon Musk himself, in February 2018, three years before he sued anyone over it.

OpenAI published the email archive on its blog in March 2024 (openai.com/index/openai-elon-musk). The receipts are in there.

The challenge is that the email archive shows Musk pushing the same direction Altman is now on trial for. To be super clear, in late 2017, Musk demanded majority equity in any for-profit conversion of OpenAI, demanded initial board control, and demanded the CEO role. When the team refused, he wrote, "Discussions are over". He added that the previous offer was "no longer on the table".

In February 2018, he forwarded an email suggesting OpenAI should "attach to Tesla as its cash cow", with his own attached commentary calling it "exactly right" and arguing that "Tesla is the only path that could even hope to hold a candle to Google".

He resigned from the board later that month.

A year later, Altman and Brockman did a version of what Musk had proposed. They created a "capped-profit" subsidiary, which is a hybrid structure where outside investors can earn a return on their money but only up to a fixed multiple, with anything above that flowing back to the nonprofit. They built it around Microsoft's $1 billion investment, with the founders carrying equity for the first time. The structures are not identical to what Musk had pushed for. But the principle he now claims he was deceived about, that OpenAI must remain a charitable nonprofit, is the principle he himself was trying to set aside in 2017. The structure worked. ChatGPT launched in 2022. The company is now valued at over $850 billion in what's called the secondary market, which is where existing OpenAI shareholders (early employees, early investors) sell their private shares to other investors before the company has formally listed on a stock exchange. The expected valuation when OpenAI does eventually list publicly is approaching $1 trillion.

Now Musk is back in court asking for reparations from the company he himself walked away from. OpenAI's lead trial attorney is William Savitt (partner at Wachtell, Lipton, Rosen & Katz, the firm that has previously represented Musk and Tesla on other matters). Savitt's opening was clean. "We're here because Mr. Musk didn't get his way at OpenAI," he told the jury, adding that Musk is "a competitor" who "will do anything he can to attack OpenAI".

The xAI distillation admission (distilling, in plain terms, is when an AI company uses another AI company's models to help train its own, getting some of the same capabilities at a fraction of the cost) gave that opening more force than Savitt could have hoped for.

The piece of evidence that turned this lawsuit into a trial is Greg Brockman's personal diary, admitted as Exhibit 43. Judge Gonzalez Rogers cited it explicitly on 15 January 2026 when she denied OpenAI's summary judgement motion (a summary judgement is when a defendant asks the judge to throw the case out before trial because the evidence is supposedly too thin to bother a jury with), and that ruling is what got the case to Oakland.

The entries follow a 6 November 2017 meeting where Brockman and Altman assured Musk that OpenAI would stay nonprofit. Brockman went home and wrote the opposite.

A few entries later, under the heading "our plan", the writing gets more explicit about the money at stake.

Brockman acknowledged in the same diary that the deception "would be pretty morally bankrupt".

When you read those entries out loud they are damning in a way that legal exhibits rarely are. The president of OpenAI is journalling, in his own hand, about deceiving his co-founder, then keeping the journal, then surrendering it in discovery.

What they prove is that OpenAI's leadership was misleading Musk in late 2017 about the structural intent of the company, and that part is unambiguous.

The trouble is what was happening on the other side of the table at the same time. Musk was emailing Altman and Brockman demanding majority equity, board control, and the CEO role in the same for-profit pivot. He wanted the same outcome. He just wanted to own it.

Not "OpenAI betrayed the charitable mission" so much as "Brockman was lying to a guy who was simultaneously trying to take the company private and put himself in charge".

That is a different story than the headline version, and it also does not make the underlying legal question disappear. Charitable trust law cares about what was promised to all the donors, including the public who saw a tax-exempt organisation soliciting funds for a stated mission. Even if Musk himself was inconsistent, donors generally are bound by the entity's representations, not the other way round. That is the question the judge actually has to rule on, and it carries weight independent of the personalities.

The dollar number Musk testified to under oath is materially smaller than the figures he has used in public for years. Musk testified that he donated $38 million to OpenAI between 2016 and 2020, across roughly 60 separate donations, including rent payments for OpenAI's Pioneer Building space in San Francisco.

That gap between the public and the sworn number cuts both ways for him. In 2023 he posted on X that he donated $100 million, and in a prior deposition he said the same. On the witness stand under oath the number was $38 million. He also testified that his original pledge was "up to $1 billion" and never delivered the rest. In his own words, he "lost confidence in the team".

The challenge is the further complication that emerged on Day 4. Jared Birchall (the manager of Musk's family office, the private operation that runs his personal wealth) testified that some of the donations went through donor-advised funds at Vanguard and Fidelity. A donor-advised fund is a charitable account where, once you put money in, you legally give up direct control over how it is spent and the fund administrator becomes the formal donor. OpenAI's attorney Bradley Wilson seized on that, questioning whether Musk had any standing to claim a personal charitable trust over money he had already legally handed off.

The $38 million figure requires more context. OpenAI's nonprofit took in about $133 million in total donations during its early years, and Musk was the single largest donor. That fact still matters. The $1 billion he pledged does not, because he did not give it.

The harder truth on timing is that Musk knew what he is now claiming as harm long before he sued. In September 2020, Musk publicly tweeted that "OpenAI is essentially captured by Microsoft".

That was three years after he left the board. By then, Microsoft had already invested billions, and the capped-profit conversion was already public. Everything Musk now claims as harm was visible on the public record.

He did not sue in 2020, or 2021, or 2022, or 2023. He filed in February 2024, withdrew that case, and refiled federally in August 2024. He added Microsoft as a defendant in November 2024.

In March 2026, Judge Gonzalez Rogers explicitly rejected Musk's "continuing violation" theory of the statute of limitations. The statute of limitations is the legal rule that says you have a limited window of time to bring a lawsuit after the harm happens; the "continuing violation" argument is a legal workaround that says the harm is still happening today, so the clock keeps resetting. The judge said no, and that ruling already narrows what Musk can recover.

Microsoft's defence counsel, Russell Cohen, leaned hard on the timing in his opening. He also pointed out that Musk had Satya Nadella's personal phone number throughout this period and never once raised an objection that Microsoft was violating donor conditions.

To be super clear, most coverage glosses over what is, in legal terms, the heaviest fact in the case.

OpenAI completed its restructure on 28 October 2025. The for-profit became OpenAI Group PBC, a Delaware Public Benefit Corporation, which is a US legal structure that lets a company pursue a public mission alongside a profit motive (Patagonia and Kickstarter run as PBCs). The nonprofit, renamed the OpenAI Foundation, retained controlling status and approximately 26% of the PBC's value. That stake is worth roughly $130 billion, making the foundation one of the largest charitable endowments on the planet, in the same league as the Bill and Melinda Gates Foundation. The Delaware Attorney General, Kathy Jennings (Delaware is the US state where most large American companies are legally registered, so its AG is the regulator who signs off on these conversions), formally approved it.

So when Musk asks the court to "unwind the for-profit conversion", he is asking it to reverse a transaction that the relevant state attorney general already approved. Not a future risk. A completed legal fact.

A judge can still rule that the transaction was wrongful and order remedies, and courts have done that before. The practical lift of unwinding a $130 billion endowment that is already structured, regulated, and operating sits in a different order of magnitude from blocking a deal before it closes.

Polymarket, an online prediction market where people bet real money on the outcomes of news events, currently gives Musk a 36% chance of winning (polymarket.com). Prediction markets are a single signal, not a verdict. The figure tracks what most legal commentators are saying and what the judge's pre-trial rulings already imply. Musk has a hard road, and the betting market knows it.

The challenge for Musk's team is that the judge has already shown her hand on three things.

That ruling is procedural and it does not make the safety question go away. A meaningful share of the AI research community shares Musk's underlying view, which is that profit-maximising structures are the wrong vehicle for AGI development. Stuart Russell, the UC Berkeley AI safety expert testifying for Musk this week, is one of them. The substantive question stays open even if the courtroom one is closed off.

She also told both sides on Day 1, before jury selection, that "there are a lot of people out there who don't like your client". That sentence applies to Musk and Altman in roughly equal measure. The judge appears to have read the room exactly right.

There is a second story running underneath the legal one, and most of the AI industry is actually watching this one.

OpenAI is going public, possibly within twelve months, at up to $1 trillion. Microsoft has more than $13 billion of equity riding on the outcome. On the day jury selection began, OpenAI and Microsoft announced a revised partnership. The new arrangement gave OpenAI freedom to work with Amazon while keeping Microsoft as primary cloud provider through 2030.

The timing was not a coincidence. Both sides are repositioning around what the courtroom will and will not deliver. Meta has already filed a letter with the California AG supporting Musk's case, and Mark Zuckerberg has texted Musk about co-bidding on OpenAI's IPO. Competitors are circling.

Musk does not need to win this trial to damage OpenAI's IPO run. He just needs the trial to happen, and every diary entry read out in court, every uncomfortable cross-examination, every screenshot of Altman with arms crossed in the gallery becomes content that lives on the internet forever.

Tim Higgins at the Wall Street Journal said it more bluntly. Musk has already won part of this. He has spent two years questioning Altman's ethics in public, and the trial gives that questioning a courtroom backdrop and a discovery file. Meanwhile, Musk is taking SpaceX public as early as June. The target valuation is $1.75 trillion, with xAI wrapped inside, and the listing beats OpenAI to the public-market window. Anthropic is reportedly weeks away from a $50 billion raise at $850–$900 billion that could see it IPO by October. All three of those listings are competing for the same pool of public-market money this year, and the trial is happening at exactly the wrong moment for OpenAI to be defending its reputation.

Three things land for me as someone who watches AI companies build, ship, and stumble. They are not specific to OpenAI, they are the durable lessons of this trial.

That does not mean Musk has no case. The Brockman diary entries carry weight as documentary evidence, the November 2017 deception sits in his own handwriting, and the charitable trust question is a serious one for nonprofit law generally.

It does mean the moral high ground in this fight is contested terrain. Musk's case is strongest on what OpenAI's leadership did. It is weakest on what Musk himself was simultaneously trying to do, what he later admitted xAI does today, and how long he waited before objecting.

Week two starts on Monday 5 May, with a live audio stream of the proceedings opening up to the public for the first time. Greg Brockman and UC Berkeley AI safety expert Stuart Russell are first up. Sam Altman is scheduled to testify but had not taken the stand at the time of writing. Microsoft CEO Satya Nadella is on the witness list for roughly an hour. Ilya Sutskever, who already gave a ten-hour deposition on 1 October 2025, is scheduled for thirty minutes live. Mira Murati will appear by videotaped deposition.

The liability phase is expected to wrap by 21 May. After that, the judge alone decides damages and remedies, with no second jury and no televised verdict.

The most likely outcome looks like this. A partial finding for Musk on the charitable trust question, modest financial relief flowing to the OpenAI Foundation, no removal of Altman or Brockman, and no unwinding of the conversion. The realistic floor for OpenAI is a settlement, an asterisk on the IPO prospectus, and a slightly lower opening valuation than they were hoping for. The realistic ceiling for Musk is a moral win without a structural one.

For everyone else, the real takeaway is structural. The two most powerful people in AI built the most consequential company of the decade together, and neither came out of the founding structure looking great. The thing they were both fighting for, from the very first board meeting, was control of a future that had not been built yet.

That is the AI industry in one sentence.

The fights worth watching are not just about the technology. They are about who gets to own the next ten years.