Three days of cross-examination in Oakland produced a $130bn lawsuit’s most awkward admissions, including that xAI trains on OpenAI’s models. The judge, not the jury, will decide.

Elon Musk took the stand in Oakland on Tuesday with a story he has been telling for two years. He had founded OpenAI in 2015, he said, to keep advanced artificial intelligence out of the hands of any single company.

Sam Altman and Greg Brockman, his former collaborators, had then quietly turned the lab into a for-profit empire, taken billions from Microsoft, and shut him out. The lawsuit he filed in 2024 was, in his framing, a corrective: a bid to restore the original nonprofit and recover what he says was stolen.

Three days later, that story looked considerably more contested than it had on Monday.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

Musk’s case had hit “some rough spots.” Musk’s own admissions on cross-examination, the judge’s repeated warnings about the scope of the dispute, and a series of pre-trial rulings narrowing the legal claims have all combined to make the world’s richest man’s case against the most valuable AI company in the world look, at least at this stage, harder to win than its initial framing implied.

The trial opened on 28 April in the federal courthouse in Oakland, before Judge Yvonne Gonzalez Rogers. A nine-person jury was seated the day before. Musk, his lawyers, OpenAI, Altman, Brockman, and Microsoft are all in the suit.

The headline damages figure is more than $130bn, though some early coverage has cited $150bn; either way the structural remedies Musk is seeking, including a partial unwinding of OpenAI’s for-profit conversion, are arguably the more consequential ask.

The procedural setup is unusual. Although a jury has been impanelled, its verdict is advisory only. Judge Gonzalez Rogers will make the final decision on liability and on remedy, and is expected to rule by mid-May.

The trial is therefore less a contest for the jury’s hearts and minds than a long, public deposition in front of the judge, who has already pruned the case before it began. She dismissed Musk’s fraud claims pre-trial and warned both sides.

Three days, several admissions

Musk was the first witness. He spent parts of three days on the stand, first under questioning from his own counsel, then under sustained cross-examination by William Savitt, OpenAI’s lead lawyer.

The first awkward moment came on the question of nonprofit commitment, the theory at the heart of the case. Savitt produced internal documents and contemporaneous communications that, in his framing, showed Musk had pushed in 2017 and 2018 for OpenAI to convert into a for-profit under his control, and had walked away from the project when that did not happen. “

You were never committed to OpenAI being a nonprofit,“ Savitt put to him, in an exchange. Musk disputed the characterisation but conceded the documents.

The second awkward moment, audible from the gallery, was Musk’s acknowledgement that xAI, his own AI company and the maker of the chatbot Grok, distils on OpenAI’s models, in effect training on the outputs of the very system he says was wrongly converted to private gain. 

The third was procedural. Savitt argued that Musk had waited too long to sue, and that key claims were filed after the relevant statute of limitations expired. Whether the judge accepts that defence is a separate question, but the timeline becomes part of the record either way.

Even before opening statements, Judge Gonzalez Rogers had reshaped the case. Her pre-trial rulings dropped Musk’s fraud claims and confined the trial to the narrower question of whether OpenAI breached charitable-trust and contract obligations when it restructured.

That makes the case less dramatic in framing, but easier to litigate, and arguably harder for Musk to win on his original theory of grand betrayal.

On day three, Gonzalez Rogers cautioned the lawyers against treating the proceedings as a referendum on AI safety or on Altman’s character. Both sides have been prone to the slippage.

Musk, on the stand, repeated long-running arguments that AI poses existential risk, an answer the judge appeared to find tangential to the legal question of whether OpenAI’s directors breached their fiduciary duties.

What’s coming next?

The trial is expected to run another two to three weeks. Altman is set to testify, as are Brockman, Microsoft chief executive Satya Nadella, and several of OpenAI’s earliest engineers. Musk’s expert witnesses, according to court documents, include the Berkeley AI researcher Stuart Russell and the Columbia Law School tax-and-nonprofit specialist David Schizer.

OpenAI is expected to call its own roster of governance and AI-safety experts, with Axios reporting that the defendants intend to put Grok’s own safety record in front of the jury.

Musk may yet recover ground. Cross-examinations of his founding partners could produce admissions of their own; the documentary record, which neither side disputes runs into thousands of pages, is broad enough to support more than one reading. The judge, not the jury, will decide, and her record so far suggests a willingness to rule on the merits rather than the theatrics.

But the first week, judged on its own, did not go well for the plaintiff. A case that began as a story about a betrayed mission has become, in places, a case about a litigant whose own conduct is now part of the evidence.

If that complicates the verdict or simply colours the coverage is something only Judge Gonzalez Rogers will settle, sometime in the next few weeks.