The charity-theft theory, on the stand

Elon Musk took the stand in Oakland on Friday in Musk v. Altman et al., day two of the trial over OpenAI's conversion from a 501(c)(3) to a public benefit corporation owned, in substantial part, by a capped-profit structure that has since been uncapped. He called it, on the record, "charity theft." His lawyers want Sam Altman removed, want disgorgement of "tens of billions" against Altman and affiliated entities, and want the conversion itself unwound.¹

I have been reading the trial filings, and the thing I want to flag at the outset is that the most consequential document in this case is not the complaint. It is the November 2024 Delaware certificate of incorporation for OpenAI Group PBC, plus the side letters governing the foundation's residual interest, plus the Microsoft amended commercial agreement that survived the conversion. Those instruments, not Musk's testimony, are what a court would have to undo. Musk's lawyers are arguing about motive. The remedy lives in the paperwork.

What is actually being asked for. Reading the second amended complaint and the proposed order, the relief sought is layered, and the layers are not equally plausible. Disgorgement against Altman personally, premised on the theory that entities he held interests in (the complaint names Worldcoin/Tools for Humanity, Helion, Retro Biosciences, and Humane among others) transacted with OpenAI on terms Altman influenced, that is a fiduciary-duty claim of a kind Delaware courts handle routinely, and it lives or dies on the transaction record. Removal of Altman as an officer and director, also conventional, also evidence-driven. Unwinding the PBC conversion and restoring the nonprofit's pre-2019 governance, that is the ask that makes corporate lawyers put their coffee down. You would have to claw back equity from Microsoft, Thrive, SoftBank, Tiger, Sequoia, Khosla, and a long tail of secondary buyers, most of whom are not parties.²

The frame that applies, and the one that doesn't. The obvious lens here is model weight lineage, the question of where the IP actually sits when an AI company restructures. The Musk theory is, in essence, that the weights, the brand, and the Microsoft relationship were all developed inside the 501(c)(3) on donated capital (Musk's $44m, plus others) for a charitable mission, and that the value created has been transferred to a for-profit vehicle whose equity is held by parties that did not fund the charitable phase. This is the carve-out question in its purest form: when a nonprofit lab spins its capability into a for-profit, what does the nonprofit retain, and at what valuation?

OpenAI's answer, on the record, is that the foundation retained a substantial equity stake in the PBC, reported in the conversion filings as roughly 25% on a fully-diluted basis, with warrants, and that this stake is itself worth more than the entire enterprise was worth in 2018.³ That argument has a certain force. It is also the argument every founder makes when they dilute a charitable funder: you're richer than you were, what are you complaining about. Whether it satisfies a Delaware Chancery analysis of charitable-trust doctrine is a different question, and one the Oakland court is not, strictly, deciding, this is a federal action sounding in fraud and RICO, with state-law fiduciary claims pendent.

The frame that I think doesn't quite fit is AI safety as market position. You could read the case as Musk arguing that OpenAI's safety posture was the charitable purpose, and that the PBC has abandoned it, making the conversion a bait-and-switch. The pleadings gesture at this. But the evidence for "abandoned safety" as a market-structural claim is thin, OpenAI has, if anything, leaned harder into safety-as-enterprise-narrative since the conversion, and the Preparedness Framework updates of the last eighteen months read as more, not less, restrictive than the pre-conversion charter commitments. Musk's safety argument is rhetorical scaffolding for the fiduciary claim. It is not the load-bearing wall.

What the primary document says. The most interesting filing in the docket, to me, is the December 2025 declaration from OpenAI's then-CFO attaching the schedule of related-party transactions between OpenAI and entities in which Altman held a 5%+ interest during 2019–2024. The schedule discloses approximately $1.9bn in cumulative transaction value across compute purchases, talent secondments, and one equity investment. The declaration argues each transaction was approved by disinterested directors and priced at arm's length. Musk's lawyers are arguing the disinterested-director process was captured. That is an evidentiary fight, not a structural one, and it will turn on board minutes and email traffic that I have not seen and you have not seen.⁴

What I would watch for, in the trial record as it unfolds, is whether any of those transactions involved transfers of model weights, fine-tuned checkpoints, or preferential API access on terms not available to third parties. That is where the model-weight-lineage frame would actually bite, not on the headline conversion, but on whether capability, separately from equity, leaked from the charitable vehicle to Altman-adjacent for-profits during the period the conversion was being negotiated. The complaint alleges this in general terms. The trial will either produce a specific instance or it won't.

What this is a case of. It is a case of late-stage governance litigation against a structure that was already legally completed and capitalised on by sophisticated counterparties. Compare the WeWork-Adam Neumann shareholder suits, which produced settlements but did not unwind the SoftBank recapitalisation. Compare, more aptly, the Hershey Trust litigation of 2002, where a charitable trust's attempt to sell its controlling stake was blocked by the Pennsylvania attorney general, except here the sale has happened, the proceeds have been spent on H100s, and the AG (California's, in this case) signed off, with conditions, in early 2025.

The realistic outcomes are, in descending order of probability: a finding for Altman on most counts with possibly some disgorgement on specific related-party transactions; a settlement in which Altman exits some affiliated positions and the foundation's stake is topped up; a finding that the conversion process was procedurally defective but with a remedy short of unwinding (additional foundation equity, governance changes, an independent monitor). The full rollback Musk is asking for would require the court to order Microsoft, SoftBank, and Thrive to return billions of dollars of equity in a company they paid for in cash, and I do not think that is what is going to happen.

What to watch. The trial schedule has Altman testifying the week of 28 April. The related-party transaction schedule will be entered as an exhibit, and once it is on PACER it will tell us more about OpenAI's actual operating relationships in 2019–2024 than four years of journalism has. I will read it the day it lands.

Footnotes

¶