Elon Musk says his xAI startup's models were partially trained on OpenAI's tech
Elon Musk continued his clash with OpenAI's attorney Thursday in Oakland federal court, with Musk saying he was being repeatedly cut off by the lawyer.
Musk's testimony during the trial's third day also revealed that his startup xAI, which is now part of SpaceX, partially "distilled" or used OpenAI's tech to train xAI's artificial intelligence. That practice appears to violate OpenAI's terms of service.
As Musk sought to prove OpenAI abandoned its nonprofit mission to improve humanity in favor of making money, he affirmed during cross-examination that there was no document saying OpenAI couldn't create a for-profit affiliate. He also said there was no document stipulating the terms of his $38 million in donations, beyond a corporate charter saying OpenAI was a nonprofit.
But he continued repeating his belief that OpenAI CEO Sam Altman and President Greg Brockman "stole a charity." Musk is alleging breach of charitable trust and unjust enrichment, seeking upwards of $134 billion in damages to fund OpenAI's nonprofit mission, and trying to oust Altman and Brockman from the company.
Musk's testimony has repeatedly referenced the prospect that AI may destroy humanity. In a Tesla earnings call last year, he said he needed more control over the company to oversee its "robot army," a reference to its Optimus robots now made in Fremont. He said both Thursday and Wednesday that he didn't mean in a military sense, but mentioned the possibility of a "Terminator" scenario.
Judge Yvonne Gonzalez Rogers said the trial is about OpenAI's nonprofit mission, not the prospect of a tech extinction event. She told Musk to stop talking about it in further testimony.
"They got it. It's enough," Gonzalez Rogers said.
Near the end of his testimony, Musk said OpenAI's lawyer William Savitt was asking a leading question - which Savitt is allowed to do during cross-examination. Gonzalez Rogers reminded the courtroom that Musk is "not a lawyer." "I did take Law 101," Musk said.
When asked if he was aware that OpenAI's nonprofit arm is funding healthcare research including Alzheimer's studies, Musk said he didn't know what OpenAI is doing.
Released evidence also pulled the curtain back on Musk's interactions with another of tech's most powerful leaders: Meta CEO Mark Zuckerberg. In a December 2024 text exchange, Zuckerberg told Musk that someone leaked a letter from Meta to California's attorney general that supported Musk's lawsuit against OpenAI.
In another text on Feb. 3, 2025, Zuckerberg praised Musk's Department of Government Efficiency project. "Looks like DOGE is making progress. I've got our teams on alert to take down content doxxing or threatening the people on your team. Let me know if there's anything else I can do to help."
Musk wrote to Zuckerberg, "Are you open to the idea of bidding on the OpenAI IP (intellectual property) with me and some others?"
Zuckerberg said "want to discuss live?" and Musk liked the text message and said he would call in the morning. Zuckerberg ultimately didn't join Musk's $97.4 billion bid that month for OpenAI, which was rebuffed.
Musk completed his testimony around 10:45 a.m. and was followed by his family office head and Neuralink CEO Jared Birchall. Further scheduled witnesses include AI safety expert and UC Berkeley professor Stuart Russell and Brockman of OpenAI.
This article originally published at Elon Musk says his xAI startup's models were partially trained on OpenAI's tech .
