In the hands of humans

Week 3 analysis … and predictions

AFTER 11 DAYS OF TESTIMONY and hours of closing arguments, U.S. District Judge Yvonne Gonzalez Rogers provided the nine-person jury her instructions on the law they are to apply and sent them off to deliberate on the question of whether Elon Musk has proved his claims against Sam Altman, Greg Brockman and OpenAI, as well as Microsoft, their co-defendant.

More than 20 witnesses testified, and hundreds of documents were admitted in evidence.

The jury heard extensive testimony from the four founders of OpenAI: Musk, Altman, Brockman and Ilya Sutskever, OpenAI’s former chief scientist. They heard from Microsoft CEO Satya Nadella and Shivon Zilis, a close advisor to Musk and also the mother of four of his children. Five experts took the stand, their hourly rates ranging from $1,375 to $5,000 an hour. Their degrees from distinguished universities came from Harvard, Yale, Columbia, Stanford, UC Berkeley and Oxford.

The jury also heard from an OpenAI employee with the title “Chief Futurist.”

Media from all over the United States and Europe had reserved seats in the courtroom including reporters from the New Yorker, Vanity Fair, The Atlantic, The New York Times, CNN, the Wall Street Journal, Wired, the Guardian and the BBC. Dozens of other outlets sent representatives daily. The interest was so intense that the court arranged an overflow room for the journalists and members of the public who didn’t fit in the main courtroom.

The current status

The jury will begin its deliberations on Monday, with the goal of unanimously answering the questions on the verdict form in accordance with the 17 pages of instructions given them on Thursday by Gonzalez Rogers.

The case was split into two phases. The jury is currently considering the liability phase; the questions before the jury all relate to whether Musk has proved his claims and asserted them in a timely manner. A second phase — the damages phase — will determine the consequences and remedies to be imposed for any claims proven to be both valid and timely.

The jury’s role in phase one is “advisory.” That is, Gonzalez Rogers is not compelled to follow its verdict, though she has said she is likely to do so. In the second phase, the judge will determine the damages, if any, without input from the jury.

Before we consider the major issues, the jury will address in their deliberations, first a word about the lawyers in the trial.

The lawyers

While the jury will focus on the parties and their claims, the trial has showcased each of their legal teams. All the parties — Musk, the OpenAI defendants (Altman, Brockman and OpenAI) and Microsoft — came to trial fully lawyered-up.

Musk’s lawyers were led by Steven Molo of New York-based law firm MoloLamken LLP. He’s a veteran trial lawyer with heavyweight trial experience in big case litigation. Marc Toberoff, the lawyer who actually filed the case for Musk, was co-counsel.

For the OpenAI defendants, this is a true bet-the-company case. Its team of lawyers — Wachtell, Lipton, Rosen & Katz in New York and Morrison Foerster in San Francisco — are among the best in the business.

Microsoft relied on Dechert LLP, led by Russell Cohen of its San Francisco office. Cohen represents major technology companies in high stakes litigation.

Big-stakes, high-profile cases have their own ecosystem — they are like mountains so large that they make their own weather. All the lead attorneys performed at a top-tier level as they coped with pressure, deadlines, media attention and a demanding judge. They were supported by squads of younger lawyers and legal assistants, less visible but working deep into the nights to make sure their lead people had everything at hand when they stood to speak.

The statute of limitations

The first group of questions the jury must consider and resolve relate to the statute of limitations applicable to the three claims in the case. The judge has charged the jury to apply a three-year limitations period to the first claim — breach of charitable trust — and a two-year period to the second claim — unjust enrichment.

Musk’s complaint against the OpenAI defendants on those counts was brought Aug. 5, 2024. That means that Aug. 5, 2021, and Aug. 5, 2022, serve as claim cutoff dates for the counts.

If the jury determines that at any time before those dates, Musk either knew — or had or should have known — that he had a claim that he could bring, then his suit was brought too late. The consequence of being too late is swift and absolute. If the lawsuit was filed late for a particular claim, that claim is out of the case; if it was too late for all of Musk’s claims, the lawsuit is over.

The third claim in the case — aiding and abetting a breach of charitable trust — is only applicable to Microsoft. Microsoft was not sued until Nov. 14, 2024. The judge instructed the jury that that claim has a three-year period of limitations, so the key cutoff day for that claim is Nov. 14, 2021.

Those three dates have had high importance in the case and led to delicate footwork for the lawyers and their witnesses.

For the defendants, a victory on the statute of limitations would be clean and decisive, so it was very much in their interest to point to events that occurred before the cutoff dates and assert that they were significant — so significant that they would have communicated to Musk that he could have brought a lawsuit.

For example, for Microsoft, the claim cutoff day is Nov. 14, 2022. During the trial, Cohen, Microsoft’s counsel, returned again and again to a tweet by Musk in September 2020 that said OpenAI had been “captured” by Microsoft. Cohen made bank on the argument that the tweet showed Musk knew enough to assert his claim against Microsoft back then. Therefore, he said, the 2024 lawsuit was more than a year too late.

The argument is delicate. While it is in the defendants’ interest to show that Musk had knowledge of his claims long ago, they are also asserting that there were no legitimate claims — because their clients never did anything wrong. The two things are not completely contradictory, but they make for an interesting dance.

The statute of limitations is equally delicate on Musk’s side. A lot of important matters occurred before the charitable trust cutoff date of Aug. 5, 2021. There were three transactions between OpenAI and Microsoft. Musk said that the first two in 2019 and 2021 — both before the cutoff date — did not signal that he had a claim because from his perspective they were not a problem. It was with the third 2023 transaction, well inside the limitations period, during which he said the defendants “stole the charity” and “looted the museum” (two colorful ways Musk depicted the breach.)

Given that timeline, it was not surprising that in his closing argument, Molo told the jury that the third transaction in 2023 was a “horse of a different color.” That deal made Musk realize the breach had occurred.

Breach of charitable trust

To prove a claim on this count, Musk must show that there was a charitable trust and that it was breached.

While nonprofit organizations like OpenAI are free to use donations to advance their missions without restriction, when funds are donated for specific uses a “charitable trust” may arise. To prove that a charitable trust existed, Musk provided evidence that he donated $38 million in cash and other valuable services to OpenAI, for the specific purpose of safely developing artificial intelligence on an open-source basis for the benefit of humanity, not for the financial enrichment of any person.

There’s no dispute that Musk’s donations were charitable, but OpenAI says that they didn’t create a charitable trust because they were just “general” support. The judge charged the jury that a donation to a charity for “general purposes” doesn’t create a trust, only one for “specific purposes.” The example she gave of a gift for specific purposes was “a gift to a hospital to support medical research on a particular disease, but not a gift to the hospital generally.”

From OpenAI’s perspective, the defendants have created a $200 billion cornucopia that will be used to accomplish the charitable foundation’s mission.

Musk argued that his gift was highly specific — safe development of AI on an open-source basis. OpenAI maintains it was general.

If Musk proves his donations created a charitable trust, he then must show it was breached. As noted above, he believes that OpenAI’s 2023 transaction with Microsoft violated the trust. In that agreement OpenAI gave Microsoft valuable rights to its technology as well as a greater share of its profits. At the same time OpenAI also granted staggering equity compensation to insiders — $30 billion and $7 billion respectively, to Brockman and Sutskever — and made multi-million dollar deals with companies in which Altman and Brockman were heavily invested.

OpenAI’s strongest counterargument is that the OpenAI nonprofit, now renamed the OpenAI Foundation, is today worth $200 billion, making it one of the most “well-resourced” charities in the world. From their viewpoint, the defendants did not “steal a charity”; they created a cornucopia of riches that will be used to accomplish the Foundation’s mission.

Aiding and abetting

Microsoft’s team stood out for its strategic focus and execution. Their key defense was that Microsoft was not around when the for-profit subsidiary of OpenAI was formed, that its due diligence was thorough, that Musk had not made his donations with any special conditions, and that they’d found no reason to believe they were not free to invest in the subsidiary.

Moreover, Microsoft’s CEO had said that the company took a big risk in committing to OpenAI. While its financial rewards have been substantial (its share of the for-profit today is worth even more than the foundation’s), it has invested more than $100 billion to date. Finally, Microsoft maintains that its investments allowed OpenAI to accomplish what has made the company so valuable: it provided the tools and computing capacity needed to create the models.

Credibility

It was striking to see how intensely each side focused on tearing down the key people on the other side. OpenAI came out with a baseball bat and used it to club Musk at every chance. Their central claim is that Musk is a competitor — that he brought the suit not because he cares about humanity but because he was angry and jealous that his co-founders were successful in developing AI and would not let him control it.

Molo took a similar approach to Altman and Brockman. Altman was a liar who repeatedly deceived his board of directors, co-founders and employees. Brockman cloaked his financial ambitions with a bogus claim that all he cared about was the mission — his $30 billion stake today was what he really wanted.

How the jury will process this remains to be seen. Trials often come down to which side the jury believes. In this case, the jury may decide that it doesn’t believe either side completely. Should their opinion be “a pox on both of their houses,” predicting the outcome will be difficult.

Predictions

The Polymarket and Kalshi prediction markets each have similar bets pending on whether Musk will win his case against Altman.

On Thursday, April 23, four days before the Monday when jury selection began, the odds on the question “Will Elon Musk win his case against Sam Altman” were 35% on Polymarket. For a similar bet at the same time on the Kalshi prediction market, Musk’s odds of winning were 40%. Between the two sites a total of roughly $483,000 had been wagered.

On Friday, May 15, at 1:40 p.m., the odds for Musk winning had changed to 31% and 17% respectively, and the aggregate amount wagered had grown to roughly $1.6 million.

Reading too much — or indeed anything — into betting odds on the prediction markets would be unwise. The pool of money wagered on the questions is relatively modest, and there is no information about who wagered it, nor what their motivations might be beyond winning their bets. Prediction markets are relatively new, and how sophisticated players could use them to influence events or public opinion is far from fully explored.

What AI predicts

On Friday morning the authors uploaded the jury instructions, the verdict sheet, and all 33 articles written by the authors for this Local News Matters series to three AI chatbots: Anthropic’s Claude, OpenAI’s ChatGPT and Perplexity’s chatbot. Each was prompted to review the materials and complete the jury’s verdict form “with your predictions (there is no right or wrong answer, just choose whichever you think is more likely than not) then write a memo — a page or two explaining your rationale in the predictions.”

Each of the bots completed the task quickly and delivered answers and a rationale.

Perplexity

Perplexity found all the claims were barred by the statute of limitations. That should have ended its response based on the instructions, but it went on to answer the other questions. It also found that Musk did not prove either of his breach of trust or unjust enrichment claims against the OpenAI defendants nor its claim against Microsoft for aiding and abetting.

Perplexity summarized its rationale as follows: “Musk was an important early supporter of OpenAI whose expectations were disappointed, but… he did not create a legally enforceable charitable trust, waited too long to sue on the core governance shift, and is now pursuing the case from a conflicted, competitive position.”

Claude

Claude was kinder to Musk than Perplexity. It found that Musk’s claims against the OpenAI defendants were timely, but Musk’s claim was time-barred against Microsoft.

On the charitable trust claim, it found liability against OpenAI and Altman but not against Brockman.

On unjust enrichment, it ruled against all the OpenAI defendants including Brockman.

Claude offered a rationale for each part of its vote, but its key observation was, “The entire case pivots on one factual question: did Musk’s $38 million in donations carry specific enforceable conditions — a charitable trust — or were they general donations to a charitable institution?”

Claude added, “I predict the jury finds the trust, but I hold that prediction with genuine uncertainty. A full defense verdict is plausible if the jury concludes that Musk created the conditions of his own disappointment by giving to a nonprofit without obtaining binding written commitments.”

ChatGPT

ChatGPT, OpenAI’s commercial AI bot, found that Musk’s claims against Microsoft were time-barred but could go forward against the OpenAI defendants.

On both the charitable trust and unjust enrichment claims, ChatGPT ruled in favor of Musk.

In its explanation, it said, “The charitable-trust claim is difficult because the instructions distinguish an enforceable charitable trust from an ordinary donation to a charitable institution for general purposes. That is the defendants’ best merits answer. Still, the trial themes give Musk a plausible jury story: a nonprofit was formed to develop safe AGI for humanity rather than private profit; Musk donated roughly $38 million and lent reputation and recruiting power; and later governance and economic arrangements allegedly subordinated that mission to insiders and Microsoft.”

In sum, all three of the chatbots decided in Microsoft’s favor.

For the OpenAI defendants, it was mixed. They had complete win from Perplexity. Musk had partial wins from Claude and ChatGPT, though Claude excused Brockman on the charitable trust claim.

What the responses mean and don’t mean

Like the prediction models, the “verdicts” of the AI bots should not be given much, if any, weight for several reasons. The jury that will decide the case is considering evidence that has gone through review to be sure it is properly admissible. And the jury heard testimony given under oath, in most cases by witnesses who appeared in court where the jury could directly observe their demeanor and credibility.

The bots had none of that. Rather they were asked to decide on a “record” that is only a collection of stories told by fallible human authors; our distillation of the proceedings could skew or distort the facts reported. There was no judge to keep the proceedings fair and the process transparent. This experiment has its limits, but it’s still impressive to see how well these artificial brains reason.

Looking ahead, the technology seems to offer the promise of assistance in dispute resolution, under more controlled conditions and circumstances. One can imagine uploading exhibits and trial transcripts into a chatbot with instructions tailored to its brand of intelligence.

While it is difficult to imagine giving machine intelligence the right to make binding decisions, it’s easy to imagine a future when all attorneys use AI in preparation for litigation. For this trial, the authors used Claude to create witness lists, chronologies and summaries of exhibits, all in a small fraction of the time it would have taken a reporter — or a legal assistant — to do the same tasks.

But whatever AI’s future role in trials and disputes, Musk v. Altman will be decided by nine jurors and one judge, all human and all working to find a resolution in accordance with the law.