Musk v. OpenAI: the founder takes the stand

Elon Musk finished three days of testimony in Musk v. OpenAI in the Northern District of California on Wednesday, in front of Judge Yvonne Gonzalez Rogers, in a courtroom in Oakland. He is asking for $150 billion in damages and a set of governance remedies that would, in effect, unwind OpenAI's conversion to a for-profit. He claims the nonprofit he co-founded was "looted" to capitalise an $800 billion private company. Cross-examination, by OpenAI's lead trial counsel William Savitt of Wachtell, was, to use the technical term, heated. At one point Judge Gonzalez Rogers told Musk to stop accusing Savitt of asking "leading" questions, which is both a thing only the judge gets to rule on and a thing Savitt was largely not doing.¹

I want to be careful here, because the trial is the show and the structure is the story, and these are not the same thing.

The show. The show is Musk on a witness stand for three days, alternately combative and aggrieved, telling a jury that Sam Altman took something he, Musk, helped build for humanity, and turned it into a personal empire. It is a good show. Musk is good at the show. The $150 billion damages number is a show number, it is roughly the implied value of the equity Musk says he would have held had the nonprofit not converted, calculated against OpenAI's most recent secondary-market valuation, and it is the kind of number you put in a complaint to get a jury's attention rather than the kind of number a court is likely to award.

The structure. The structure is more interesting and is what I'd actually watch.

What Musk is litigating, stripped of the personal grievance, is whether you can take charitable assets, money donated to a 501(c)(3), research produced under a charitable mission, and crucially the model weights and research outputs developed during the nonprofit period, and roll them into a capped-profit and then uncapped-profit structure without compensating the charity at fair value. This is, as a matter of California nonprofit law, a real question. The California Attorney General has been circling it independently, and Delaware's AG made noises during the conversion process last year. Musk is, in effect, running a private enforcement action on a question the state AGs have been slow to bring themselves.

The remedy he wants, beyond the damages, is the structural piece: an injunction either reversing the conversion or requiring OpenAI to compensate the nonprofit foundation at a value reflecting what the for-profit is now worth. The latter is the realistic ask. If granted, it would require OpenAI to issue a very large block of equity, or pay a very large cash sum, to a foundation whose board Musk would like to see reconstituted. The damages number is theatre; the equity transfer is the actual prize.

This is where the model weight lineage frame earns its keep. The thing the nonprofit produced, and the thing the for-profit now monetises, is a lineage of model weights, GPT-2, GPT-3, GPT-4, and onwards, each trained on top of the prior, each a derivative work in a technical sense even where it isn't in a legal one. The legal system does not yet have a clean theory of how to value weights that were produced under one ownership structure and now sit inside another. Patents have a transfer regime; trade secrets have a transfer regime; weights, which are neither exactly, do not. Musk's case is, among other things, a forcing function on that question. If the jury or the judge has to put a number on what the nonprofit's research contributed to the for-profit's current value, somebody has to value the weights. I would read that valuation testimony carefully when it comes; it will be cited in every AI-company carve-out for the next decade.²

The safety-as-market-position frame applies less directly but is in the room. Part of Musk's narrative is that the nonprofit's mission was safety-first AI development, and that the for-profit conversion subordinated safety to commercial pressure, and that this constitutes a breach of the charitable purpose. OpenAI's defence on this is that its current structure preserves the mission through the nonprofit's continued board control of the for-profit, that its Preparedness Framework and RSP-equivalent governance are in place, and that the conversion was necessary to raise the capital required to pursue the mission at all. The defence is, in essence, that safety posture is preserved by structure even after conversion. Whether the jury believes that is a market-relevant question, because OpenAI sells against Anthropic partly on the basis that its safety governance is comparable, and a finding that the conversion gutted the mission is the kind of thing enterprise buyers in regulated sectors read.³

A few things worth noticing about the cross-examination, based on the reporting:

Savitt apparently spent meaningful time on Musk's 2017–2018 emails proposing that he, Musk, take majority control of OpenAI and merge it into Tesla. The point of this line is not to embarrass Musk but to undermine the breach-of-fiduciary-duty theory: if Musk himself proposed converting the nonprofit's work into a for-profit structure under his ownership, his standing to complain that someone else did roughly the same thing under different ownership is, let us say, complicated. This is a strong line for OpenAI and I'd expect it to feature heavily in closings.

Musk's own xAI is now valued in the tens of billions and competes directly with OpenAI. Savitt would be derelict not to put this in front of the jury, and reportedly has. The competing-interest framing turns the case from "founder protecting charitable mission" into "competitor using nonprofit law as a strategic weapon", which is a much harder narrative to win a jury with.

The judge's "leading questions" rebuke is a small thing, but small things in front of juries are not always small. Witnesses who fight the lawyer fight the judge, and witnesses who fight the judge lose juries. Musk has been, by his own brand, combative. Whether that reads as conviction or as the thing OpenAI's defence says it is, a personal vendetta dressed in charitable-trust law, is partly a function of how Musk handles the next two weeks.

What to watch. The valuation phase, if the case gets there: who OpenAI puts up to value the weights, who Musk puts up, and what methodology survives. The California AG's posture: a state intervention, even a late one, would change the case's centre of gravity. And the settlement temperature: cases like this often settle on the courthouse steps when the equity-transfer remedy starts to look real to the defendant. I would not assume a verdict.

Footnotes

¶