Musk Just Admitted xAI Trained Grok on OpenAI’s Outputs—While Suing OpenAI for Going Commercial
Musk admitted xAI distilled OpenAI outputs to train Grok—while suing OpenAI for abandoning its humanitarian mission. The $134B lawsuit is getting awkward.
- Musk admitted xAI “partly” used OpenAI’s outputs to train Grok—OpenAI’s terms of service explicitly prohibit that.
- The man suing OpenAI for abandoning its humanitarian mission acknowledged his rival startup benefited from OpenAI’s work.
- The lawsuit is seeking $134 billion to fund OpenAI’s nonprofit mission—Musk tried to buy the whole thing for $97.4 billion first.
In a courtroom in Oakland, California, on Thursday, Elon Musk admitted his AI company xAI had used OpenAI’s technology to train Grok. Not a great look for someone suing OpenAI for abandoning its humanitarian mission.
During cross-examination by OpenAI’s attorneys, Musk confirmed under oath that xAI had “partly” distilled outputs from OpenAI’s models to improve its own, according to The Verge and multiple other outlets. Model distillation—using the outputs of a larger, costlier model to train a smaller, cheaper one—is standard practice across the industry. It’s also, notably, explicitly prohibited by OpenAI’s own terms of service.
“Generally A.I. companies distill other A.I. companies,” Musk said in his testimony. The admission came as he was being questioned about xAI’s training practices in the lawsuit he filed against OpenAI, Sam Altman, Greg Brockman, and Microsoft—a case that seeks $134 billion to enforce OpenAI’s original nonprofit charter.
The Irony Wasn’t Lost on Anyone
Musk’s lawsuit argues OpenAI broke its founding promise to develop artificial general intelligence for humanity’s benefit, not for profit. But the man making that argument acknowledged in open court that his own AI company used OpenAI’s outputs to train Grok—outputs OpenAI generated under that same nonprofit mission. OpenAI banned accounts for suspected distillation earlier this year when it accused Chinese AI firm DeepSeek of the same practice.
The trial, before Judge Yvonne Gonzalez Rogers in U.S. District Court for the Northern District of California, has been oscillating between substantive governance questions and procedural theater. Rogers has repeatedly told Musk to stop invoking AI existential risk during testimony, ruling the case is about OpenAI’s nonprofit obligations, not whether the technology might destroy humanity. “I did take Law 101,” Musk told the judge at one point, after she reminded him he is “not a lawyer” when he objected to cross-examination tactics.
The trial opened last week with only two of Musk’s original claims surviving dismissal. Released text messages showed Musk asked Meta CEO Mark Zuckerberg in February 2025 whether he’d “be open to bidding on the OpenAI IP with me and some others.” Zuckerberg said “want to discuss live?” Musk liked the message and said he’d call in the morning. Zuckerberg ultimately didn’t join—Musk’s $97.4 billion bid for OpenAI was submitted alone and summarily rejected.
The damages figure—$134 billion—is not arbitrary. Musk is seeking that amount to be directed toward funding OpenAI’s nonprofit mission, effectively making his lawsuit a hostile takeover dressed as a charitable enforcement action. Altman took the stand the day before Musk and called the damages claim a “tax deduction.”
Musk completed his testimony around 10:45 a.m. on Thursday. Upcoming witnesses include AI safety researcher Stuart Russell and OpenAI president Greg Brockman.
