The Musk v. Altman trial closed last Wednesday. The jury deliberates Monday. Judge Yvonne Gonzalez Rogers will make the binding ruling — the jury's verdict is advisory.
Three weeks ago we wrote that the trial was asking the wrong question. We argued that whatever the courtroom decided, the structural question of who governs the infrastructure of intelligence would remain on no one's docket. We also wrote, deliberately and explicitly, that we did not have a side in the Musk versus Altman dispute. The problem was the architecture, not the man.
That position was honest at the time. It is no longer sufficient.
After three weeks of testimony and a careful look at what each principal has built around himself, an honest reading of the public record requires us to say what the evenhanded framing papers over: the architectures these two men have built around themselves are not equivalent. One is failing badly within institutional norms. The other is dismantling the norms themselves. The structural critique still applies to both. The trajectories do not.
This is not a reversal. It is an update on evidence. We owe the reader the explicit acknowledgement that we are saying something now that we did not say three weeks ago, and that the reason is what the witnesses said under oath in the federal courthouse in Oakland.
What the Trial Actually Established About Sam Altman
Strip away the personality theater and the federal record contains five corroborating witnesses to a documented pattern of dishonesty at the top of the most-watched AI company in the world.
Ilya Sutskever, OpenAI's co-founder and former chief scientist, testified that he spent months gathering documented evidence of what he described as a "consistent pattern of lying." He described his motivation as "a great deal of ownership" over the startup and not wanting it "to be destroyed." That is the structurally hardest testimony to dismiss. It comes from the person who built the technology. He had no commercial incentive to harm Altman. He documented behavior over time before going to the board.
Mira Murati, former chief technology officer, testified to resistance to board oversight and dishonesty with senior leadership. Helen Toner and Tasha McCauley, former board members, corroborated from board-side observations. Musk's lawyer's closing metaphor — "Would you cross a wooden bridge over a deep gorge if it were built on Sam Altman's version of the truth?" — landed because the corroboration is broad and source-diverse.
The cross-examination on self-dealing involving companies that do business with OpenAI was not directly denied. The defense pivoted to attacking Musk's standing instead.
The headline reading of this is "Altman is a liar." That reading is supported by the testimony but it is also incomplete. The more accurate reading is structural: this is what happens when a founder who likely began with mission-aligned intent is asked to pace a multi-hundred-billion-dollar capital build against well-funded competitors operating without the constraint of an original charitable trust. Sutskever's "I cared for it and I didn't want it to be destroyed" is not the language of a vindictive colleague. It is the language of someone who watched the founder he co-built with become someone else over time.
Strict honesty requires saying: pure nonprofit governance at the frontier of general-purpose AI development, with the current economics of compute and talent, probably cannot pace the field. The choice of "stay nonprofit and stay competitive" is most likely not actually available. The choice is between governance structures that can absorb commercial capital while preserving mission and structures that cannot. OpenAI tried to thread it with the capped-profit subsidiary model. The cap kept moving. The board got captured. The October 2025 recapitalization at $500 billion is the on-record evidence of how that thread failed.
This matters because it would be intellectually dishonest to frame Altman as a unique villain in a story where the structural pressures push every founder in his position toward roughly the same set of decisions. Anthropic is threading the same constraint with a Public Benefit Corporation structure plus a Long-Term Benefit Trust, and threading it more successfully so far. The third path exists. OpenAI did not take it.
The question the trial is asking is whether Altman crossed a specific legal line on the 2025 conversion. The question worth asking is why the available paths through this constraint are so narrow, and why the most-resourced lab in the world chose the one most likely to corrupt itself.
What the Public Record Establishes About Elon Musk
The defense did real damage on Musk's side of the case. The "selective amnesia" charge landed — he supported various OpenAI commercial moves during his board tenure and after his departure before publicly opposing them. He skipped portions of his own trial for a trip to China, alongside other ultra-high-net-worth principals, and his lawyer apologized to the jury for his absence. Altman's attorney pointed out that people close to Musk, including the mother of his children, "can't back his story."
Those are the courtroom facts. They damage his standing as plaintiff. They do not damage what is on the broader public record about the trajectory he has built outside the courtroom.
That broader record now includes, with citations and dates:
SpaceX worker safety. A Reuters investigation in 2023 documented more than six hundred injuries between 2014 and 2022 — amputations, broken backs, electrocutions, and one death. Internal pattern of underreporting OSHA-recordable incidents. Non-disclosure agreements pressed on injured workers. This is not historic. It is the documented operating culture of the company that is also seeking regulatory approval for unprecedented orbital infrastructure.
Neuralink animal welfare. USDA investigations and reporting from the Physicians Committee for Responsible Medicine documented allegations of monkey deaths from rushed experiments. The pattern: regulatory frameworks treated as friction, ethical review treated as obstacle to ship velocity.
Confirmed drug use. The Wall Street Journal's reporting on ketamine, cocaine, ecstasy, and LSD use was not effectively contested. The drugs themselves are not the load-bearing point. The load-bearing point is that the person making unilateral decisions about communications infrastructure used in active war zones, about AI-accelerated orbital compute, about the dismantling of federal agencies, and about the direction of a company seeking to put chips in human brains is operating under cognitive conditions he himself has acknowledged are non-baseline.
Estrangement from his own children. Multiple public estrangements, including Vivian Wilson's on-the-record disavowal. The family-credibility attack at trial — that even people close to him "can't back his story" — was not out of nowhere. It is a pattern documented in his children's own statements.
Unilateral private control of conflict-zone communications. Walter Isaacson's biography established, and Musk himself confirmed, that during a 2022 Ukrainian operation against the Russian Black Sea fleet, Musk refused to extend Starlink coverage over Crimea — a private decision by a private citizen that materially affected the outcome of a sovereign nation's military action against an invader. He has activated Starlink for Iranian protesters during the Mahsa Amini protests. He offered Starlink to Gaza humanitarian organizations and then negotiated terms with Israel about which territories the service would cover. Reasonable people can disagree about whether each of these decisions was right. Reasonable people cannot disagree that a private citizen is making them, unilaterally, with no accountability to the populations affected or to the international system that is supposed to govern communications during armed conflict.
The dismantling of US foreign-aid and global-health infrastructure under DOGE. This is the heaviest line on the page and it requires saying carefully. Under Musk's leadership of the Department of Government Efficiency during 2025, USAID was dismantled — ninety percent of contracts canceled by July 2025. The President's Emergency Plan for AIDS Relief, which had supported antiretroviral therapy for roughly twenty million people globally, was placed under a ninety-day pause that closed clinics across sub-Saharan Africa within days. Modeling by ImpactCounter estimated approximately 159,000 adult deaths in sub-Saharan Africa attributable to that ninety-day pause alone. Modeling published in The Lancet projected one million additional pediatric HIV infections, 500,000 additional pediatric AIDS deaths, and 2.8 million additional orphans over the subsequent five years. UNAIDS lost 55% of its head office staff. The World Health Organization was formally withdrawn from by the United States in January 2025. Bill Gates, who is not given to dramatic public framing, accused Musk in May 2025 of "killing children." We are using neutral language here only because the documented record does not require dramatic language to make the point. The numbers do.
That is the public record on the principal who is, in federal court this month, the named plaintiff in a charitable-trust lawsuit alleging breach of a public-interest commitment.
We wrote in April that this was the piece of the puzzle no one in the courtroom was talking about. It is still the piece no one in the courtroom is talking about. The accumulation has accelerated since.
Why the Asymmetry Matters
The mainstream framing of this trial has been "two flawed billionaires." That framing is wrong in a specific way: it suggests that the harms are roughly comparable and the structural critique applies symmetrically. After three weeks of testimony and a year of documented public conduct, that suggestion does not survive contact with the record.
Altman's pattern is corrosive within institutions. He bends and breaks rules inside an institutional architecture he is, on balance, trying to remain inside of. The harms are concentrated in one vertical — AI governance norms — and they are serious. The corrective is structural reform of AI governance specifically. It is a hard problem. It is a bounded problem.
Musk's pattern is dismantling. He is not navigating institutional constraints. He is, as a matter of documented public conduct, trying to make the constraints stop applying. The harms are distributed across critical infrastructure layers: physical safety of workers, animal welfare in research, global communications during armed conflict, public health systems abroad, federal agency capacity at home, and now the explicit request for compute infrastructure operating outside national jurisdiction. The corrective is not reform of a sector. There is no sector that contains the trajectory.
This is the load-bearing distinction. Both men have damaged trust. Only one of them is, as a matter of stated intent and accelerating action, building a private trajectory in critical infrastructure that he has explicitly framed as one no terrestrial regulator can reach.
The trial cannot litigate this distinction. The trial is about a 2015 charitable trust and a 2025 corporate restructure. But the public record is the public record, and any honest analysis of the moment has to name what is in it.
Prediction
The jury's verdict is advisory. Judge Gonzalez Rogers makes the binding ruling.
Our read on the jury: a partial finding of breach is the most likely outcome — roughly 55 percent probability. The five-witness credibility testimony is durable; the standing damage to Musk is also durable. Jurors who weigh evidence will not award everything Musk asked for but will not dismiss the breach claim entirely either. No finding sits at roughly 30 percent. Strong finding in Musk's favor sits at roughly 15 percent.
Our read on Judge Gonzalez Rogers: limited finding of partial breach with a modest restitutional remedy is the most likely outcome — roughly 50 percent. A defined percentage of the recapitalization value redirected to the original nonprofit foundation. No removal of Altman. No unwinding of the October 2025 recapitalization. No-breach ruling sits at roughly 30 percent. Strong breach with material remedy sits at roughly 15 percent. Full unwinding of the recapitalization — what Musk has formally requested — sits at roughly 5 percent.
We expect the judge to be cautious about industry-wide precedent and aware that a strong ruling for Musk on the charitable-trust theory creates the possibility of an xAI acquisition of OpenAI's nonprofit foundation assets at fire-sale terms. That second-order risk argues for a modest ruling that punishes the breach without enabling the broader concentration.
What the Trial Will Not Settle
It will not settle whether AGI development under for-profit governance is structurally sound.
It will not settle whether the recapitalization template OpenAI executed will be available to other labs.
It will not settle who should be at the table when decisions about transformative AI deployment are made.
It will not settle how charitable trust law adapts to entities whose charitable purpose is the development of technology that may exceed the cognitive capacity of the donors and trustees who funded it.
It will not settle whether private control of communications infrastructure in active conflict zones is compatible with the international system of state sovereignty.
It will not settle whether the dismantling of US global-health infrastructure during 2025 — with its measurable cost in lives — was a permitted exercise of executive efficiency or an unprecedented privatization of public obligation.
It will not settle whether any private actor should be permitted to seek regulatory approval for compute infrastructure operating outside national jurisdiction.
The advisory jury and the binding judge are answering a narrow contractual question about a specific 2025 restructure. They are not answering the architectural question about what governance for systems of this kind should look like. That question has been ducked by Congress, ducked by international bodies, ducked by the industry itself, and now ducked by the courts as well — not because the courts are unwilling, but because the case before them was not built to ask it.
The Line We Will Not Soften
The cost of one principal being wrong, even once, on the trajectory we are describing is not a cost paid by the principal. It is paid by people who never had standing to consent to his judgment. Workers on factory floors in Texas. Monkeys in research labs. HIV patients in Lesotho and Uganda and Mozambique. Civilians under fire in cities whose internet access depends on a billionaire's mood. Children orphaned because a ninety-day funding pause closed the clinics their parents depended on.
This is not a private mistake working itself out in a private system. It is a global externality with a measurable death toll already on the record and no enforcement mechanism that touches it.
People are entitled to draw their own conclusions about what to call that. We will use the most neutral word the public record supports: reckless. Documented patterns of reckless conduct, by a single principal, accumulating across the infrastructure layers that determine outcomes for everyone else. The trial does not address it. No regulator addresses it. The international system was not built for it.
That gap is the work.
What Happens After
Watch two specific signals after the verdict.
Inside OpenAI: whether senior alignment, safety, or governance staff begin departing in the sixty to ninety days following the ruling. That is the internal verdict — independent of what the court decides. The court can find no breach and the technical staff can still walk. The court can find significant breach and the staff can still stay if the political settlement holds. The departure pattern is the durable signal.
Across the industry: whether other labs adjust their governance posture. Anthropic, DeepMind, the broader field — the trial creates a public reference point for what charitable-trust violation looks like in the AI sector. Other labs with charitable-phase commitments will be re-reading their own founding documents this week. Some will reinforce structural protections. Some will accelerate quiet conversions before precedent hardens. And xAI, regardless of the verdict, will move on whatever opening the next thirty days create. If OpenAI's governance fractures from the ruling, the acquisition logic does not require a strong court finding — it requires a market signal that the institution is not stable.
The trial is one moment in a longer arc. It will produce a verdict and a ruling. It will not produce an architecture.
That is the work that remains. We were founded to do it. We continue.