Elon Musk Admits xAI Used OpenAI’s Models to Train Grok

So much for the principled crusade.

Elon Musk was on the stand in a California federal courtroom Thursday, and under oath he admitted something that makes his ongoing lawsuit against OpenAI look a bit… awkward. His own AI startup, xAI, used OpenAI’s models to train Grok.

The practice in question is model distillation. It’s a common technique where a larger, more capable model (the “teacher”) helps train a smaller one (the “student”). Done internally, it’s fine—Meta does it, Google does it, everyone does it. But when smaller labs use a competitor’s model this way, it gets murky. That’s exactly what Musk’s xAI did.

When asked if he knew what model distillation was, Musk said yes—and then confirmed xAI used OpenAI’s models to improve Grok. The irony is thick enough to cut with a GPU. Musk is suing OpenAI for allegedly abandoning its nonprofit mission and becoming a closed, profit-driven entity. Meanwhile, his own company was quietly using OpenAI’s work to boost its own model.

Look, I’m not saying distillation is inherently wrong. It’s a tool. But when you’re building a legal case around the idea that OpenAI is being unfair and secretive, and then you admit to using their models to train your product, you lose the moral high ground. Fast.

Musk has been positioning xAI as the open, transparent alternative to OpenAI. Grok was supposed to be the edgy, unfiltered chatbot that speaks truth to power. Turns out it needed a little help from the very power it was speaking truth to.

This isn’t just a courtroom embarrassment—it’s a strategic headache. If the court decides distillation of a competitor’s model without permission is a problem, Musk just handed OpenAI ammunition. If it decides it’s fair game, then OpenAI’s case against open-source model use gets weaker. Either way, Musk’s testimony doesn’t help his narrative.

What’s really funny is that Musk has been loudly criticizing AI labs for using data without permission. He’s called out Google, Microsoft, and especially OpenAI for scraping and training on public data. Now we find out his own team was doing the same thing—just at the model level instead of the data level.

I don’t expect this to change much in the grand scheme of things. Distillation is widespread, and most companies will keep doing it until someone successfully sues them into stopping. But for Musk, who loves to play the victim and the reformer, this is a bad look. He can’t sue OpenAI for being closed while secretly using their open models to catch up.

We’ll see how this plays out. But for now, the guy who said he wanted to “make AI safe for humanity” just admitted he needed a little help from the company he says is making it unsafe.