Elon Musk spent week one of his federal lawsuit against OpenAI accusing Sam Altman of betraying the nonprofit mission. Then on Thursday afternoon, under cross-examination, he admitted xAI trained Grok partly by distilling OpenAI's models. Reporters in the courtroom described audible gasps. Judge Yvonne Gonzalez Rogers did not look amused.

OpenAI counsel Jordan Eth pulled the admission out in two questions. Did xAI use distillation on OpenAI models to train Grok? Musk: "Partly, yes." Was that common practice? Musk: "Generally all AI companies do that."

That sentence is going to follow xAI into its S-1.

The Hypocrisy Problem

Musk's entire complaint rests on the claim that OpenAI abandoned its founding charter by going commercial and locking up its weights behind paid APIs. He wants the jury to believe that when Altman flipped the switch from research lab to product company, he stole something Musk helped pay for. Musk's $38 million in early funding, the argument goes, seeded what is now a roughly $800 billion enterprise that the public can no longer freely use.

Fine. That is a coherent grievance. The problem is that xAI exists, raised at a $200 billion valuation, is about to IPO inside SpaceX at a $1.75 trillion blended mark, and according to its own founder it learned to talk by copying the very company he says abandoned its mission. You cannot simultaneously argue the public was robbed of OpenAI's outputs and concede that you helped yourself to those outputs to bootstrap your own product.

Distillation, for the uninitiated, is the practice of training a smaller model on the responses of a larger one. It is faster than training from scratch and dramatically cheaper. It is also against OpenAI's terms of service. OpenAI has spent the last eighteen months publicly accusing Chinese labs of doing it. DeepSeek caught the brunt of that campaign last year. The State Department briefings, the export control hearings, the entire China hawk narrative leaned on the idea that Chinese frontier models were derivative because they had been distilled from American ones.

Musk just torched that narrative under oath. "Generally all AI companies do that." If the founder of xAI says distillation is industry standard, the case against DeepSeek collapses into a debate about which laws of which jurisdiction apply, not whether the technique itself is legitimate.

What Else Happened This Week

The trial opened Monday with Musk on the stand for almost three full days. He walked the jury through what he called the "three phases" of losing trust in Altman. He warned the courtroom that AI "could kill us all," prompting Judge Rogers to interrupt and remind everyone that this is not a referendum on artificial general intelligence. Her exact line, captured by the CNBC live blog: "This is not a trial on whether or not artificial intelligence has damaged humanity." The pool reporter said she looked tired.

Day four belonged to Jared Birchall, Musk's longtime money manager, who walked through the original $38 million in tranches and the side letters that supposedly preserved Musk's nonprofit conditions. Greg Brockman was on a 48 hour testify notice for Friday. He never took the stand. OpenAI counsel asked the court to defer his testimony to next week, citing scheduling. Brockman's 2017 diary entry, the one that allegedly captures the room where Altman pitched the for profit conversion, is still pending.

The jury is now in recess until Monday.

The Money Question Just Got Ugly

xAI is the most expensive startup on the planet that nobody can value. Morgan Stanley and Goldman are circling a $1.75 trillion mark inside SpaceX. The pitch deck leans on Grok 4's benchmarks, the Memphis cluster, and the X data feed. None of those documents currently disclose that the founder told a federal jury under oath that the model was partly trained by copying a competitor in violation of that competitor's terms of service. The next S-1 will have to.

That is not a hypothetical legal exposure. It is a written admission, on the record, that creates a private right of action for OpenAI against xAI the moment this trial ends. Even if Musk wins his case, his company just handed the defendant a counterclaim with documentary support. OpenAI's lawyers know it. Their faces during the cross apparently confirmed it.

The prediction markets are not impressed. Kalshi has Musk-wins drifting at 47 percent. Polymarket has it at 42. Both moved down by mid-afternoon Thursday. The smart money read the same transcript everyone else did and concluded that the plaintiff just impeached his own moral case.

What the Jury Is Going to Hear Next Week

Brockman is the witness OpenAI wants on the stand and Musk wants to discredit. He was in the room for the conversion pitch. He took notes. Those notes are now under subpoena. If they show what OpenAI says they show, the founding mission was discussed openly and Musk signed off on a version of the for profit structure he is now suing to unwind. If they show what Musk says they show, Altman misrepresented the future direction of the company in 2017 and built a $800 billion business on that lie.

Either way the diary is the case. Without it, this is two billionaires arguing about a coffee meeting. With it, a jury has something to read.

Final Take

The cleanest summary of week one is that Elon Musk spent four days arguing OpenAI broke its promises and then admitted his own company built its product by copying theirs. That is not a winning trial strategy. It is a podcast moment that will be replayed at every AI policy conference for the next decade. Whatever the verdict, the distillation admission rewrote the rules of the conversation. Every Western frontier lab is now on the record that the technique they have spent two years denouncing is, in their founder's own words, what everyone does.

DeepSeek's lawyers are sending Musk a thank you note.