The Musk v. OpenAI trial is an evidentiary event, not an enforcement event

Jury selection in Musk v. Altman et al. began this morning in the U.S. District Court for the Northern District of California, before Judge Yvonne Gonzalez Rogers.¹ The case has been pending in some form since February 2024 and has been amended often enough that the operative complaint barely resembles the original. What goes to trial today is narrower than what was filed: in the days before jury selection, Musk voluntarily dropped his fraud and RICO claims, and the judge dismissed them at his request.² What remains are the equitable claims, breach of charitable trust and unjust enrichment, seeking up to $150 billion in damages and the removal of Sam Altman and Greg Brockman from OpenAI.¹

The first thing to understand about this trial is that the jury cannot do most of what the headlines imply it can do. Because the surviving claims are equitable rather than legal, the jury is advisory. Judge Gonzalez Rogers retains final decision authority. A jury verdict for Musk does not unwind OpenAI's for-profit conversion, does not remove Altman, and does not produce a damages award. It produces a recommendation, which the judge can adopt, modify, or ignore.³

The first thing to understand about this trial is that the jury cannot do most of what the headlines imply it can do.

This matters for how to read the event. The trial is not really an enforcement mechanism. It is a discovery mechanism with a verdict attached.

What was actually dropped, and what it tells you

Dropping fraud and RICO weeks before trial is the analytically interesting move. Fraud is jury-sympathetic, juries like to punish liars, but it requires showing scienter: that the defendant knew the representation was false when made. RICO requires a pattern of predicate acts and an enterprise, and federal judges are notoriously hostile to civil RICO theories that look like dressed-up breach-of-contract claims.

Charitable-trust breach is the opposite shape of claim. It is unsympathetic to juries (which is partly why it is being tried to an advisory one), evidentiarily it leans on documents rather than intent, and the heavy lifting is done by the founding instruments themselves, the 2015 incorporation papers, the early mission language, the donor solicitations Musk responded to with roughly $44–45 million in contributions before he left the board in 2018.⁴

The shift from "Altman lied to me" to "the charity's assets are being diverted from their stated purpose" is a shift from a hard claim to prove to a claim where the documents largely speak for themselves. I read this as Musk's counsel concluding that the structural argument is the strongest ground and the rest was noise. This is not the move of a plaintiff in retreat. It is the move of a plaintiff narrowing to the part of the case that survives summary scrutiny.

Whether it wins is a different question. OpenAI's defence, that California charitable-trust law constrains specific restricted assets rather than governance decisions about how a nonprofit restructures, and that the post-conversion PBC retains a nonprofit controlling entity with a reported ~30% equity stake, is not a frivolous defence.⁴ Gonzalez Rogers has dismissed several of Musk's earlier theories at the pleading stage, which is at least a hint that she is not predisposed.

The witness list is where the structural risk lives

The trial witness list, as reported, includes Sam Altman, Greg Brockman, Satya Nadella, and Mira Murati.⁴ Microsoft is not a defendant. Nadella is a witness.

This is the part of the case I would watch most closely if I held OpenAI exposure, directly or through Microsoft. Microsoft has put roughly $13 billion into OpenAI across tranches, structured in ways that have never been fully disclosed but that are widely understood to involve compute commitments, revenue-share mechanics, and IP licences contingent on certain corporate states. If Nadella testifies under oath about what Microsoft knew about the for-profit conversion timeline, what Microsoft's investment terms anticipated about the nonprofit's transformation, and how board-level discussions at Microsoft treated the charitable-mission language, that testimony goes into a public record that the California Attorney General, the Delaware authorities reviewing the conversion, and any future IPO underwriter will read.

Nadella does not need to be a defendant for his testimony to create exposure. The deposition transcripts, if admitted, are the asset.

Murati is the other interesting witness. She left OpenAI as CTO in October 2024 and has been publicly quiet since. She is on neither side's payroll and has no obvious incentive to align her testimony with either narrative. What she says about internal discussions of the mission and the conversion will land harder than what Altman or Brockman say, simply because she has less to gain from saying it.

The frame: this is ai-performativity risk, not governance theatre

The analytical frame I'd apply here is ai-performativity. OpenAI's $300 billion valuation and the SoftBank-led $40 billion round closed earlier in 2025 are themselves market-shaping acts.⁴ They are predicated on the for-profit conversion completing cleanly enough that a public offering becomes possible. The IPO is the exit that makes the round mark defensible.

A trial that runs into Q3 2026, with admitted documents that look bad, with a Gonzalez Rogers opinion that validates any part of the charitable-trust theory, with state-AG review proceedings drawing on the trial record, any of these creates legal cloud. Underwriters do not sign S-1s under legal cloud. They wait. Waiting compresses the window in which the $300 billion mark holds without a market test.

This is also where SoftBank's position becomes structurally relevant. SoftBank holds a large OpenAI stake and has its own capital-markets calendar, including planned partial monetisations that are at least partly contingent on OpenAI achieving public-market liquidity. A delay in the OpenAI IPO timeline transmits into SoftBank's liquidity planning. The market-flow story does not stop at OpenAI's cap table.

What this is a case of

This is a case of charitable-form companies converting to for-profit form at scales the charitable-trust doctrine was not built to handle. Anthropic is a PBC from the start. xAI is a PBC from the start. OpenAI is the only frontier lab attempting to thread the needle of converting a fully-formed nonprofit, with billions of dollars of charitable mission language in its founding instruments, into a for-profit at a $300 billion mark. There is no clean precedent for this. The trial is, in part, the precedent being made.

A reading from Gonzalez Rogers, even an advisory one, even one that doesn't enforce, establishes the doctrinal frame future AI nonprofit conversions will be structured against. If the opinion contains language about what charitable-trust obligations attach to mission-stated AI nonprofits, every future structurer reads that language.

What to watch

• Whether Nadella testifies in person or by deposition designation. In-person testimony is harder to control and produces sharper record.
• Documents admitted into evidence, especially the 2015–2018 board minutes and donor solicitations. This is where the charitable-mission language sits.
• Any interim rulings from Gonzalez Rogers on the scope of the charitable-trust theory. A pre-verdict legal ruling on doctrine is more important than the advisory verdict itself.
• OpenAI's S-1 timeline. Any signal of underwriter hesitation, even informal, is the IPO-risk channel transmitting.
• The California AG's posture. State enforcement is the actual enforcement mechanism the trial record feeds into. The judge can ignore the jury. The AG cannot ignore admitted evidence.

A $150 billion damages claim against a company with roughly $3.7 billion of 2024 revenue is not a collection target. It is a publicity and pressure instrument, and Musk, who runs xAI in direct competition with OpenAI, benefits from the pressure regardless of the verdict. That doesn't make the legal theory wrong. It does mean the plaintiff's interests and the doctrinal question are not the same thing, and a careful reader holds both at once.

Footnotes

¶