Musk Says xAI Used OpenAI Outputs to Train Grok

Elon Musk’s testimony appears to have cracked open one of artificial intelligence’s most closely guarded practices: using the work of powerful models to train new ones.

According to reports, Musk testified that xAI trained Grok on OpenAI models, a disclosure that lands at the center of a fierce industry fight over so-called distillation. The practice has become a flashpoint as leading AI labs race to build stronger systems while trying to stop rivals from borrowing, mimicking, or compressing their capabilities into smaller competing models.

The battle over AI now turns not just on who builds the best model first, but on who can stop others from learning from it.

The significance goes well beyond one company or one chatbot. Distillation promises speed and efficiency, letting developers transfer capabilities from a large model into a smaller one that runs cheaper and faster. But that same logic alarms frontier labs, which see it as a direct threat to the huge costs and technical advantages behind their systems. Musk’s reported acknowledgment gives that tension a public face.

The broader stakes look commercial, legal, and strategic all at once. If major labs rely on each other’s outputs, the industry’s competitive lines grow blurrier and the argument over what counts as fair use grows sharper. Sources suggest this issue already sits near the heart of how AI companies defend their technology, police access, and frame their rivals’ behavior.

What happens next will matter far beyond Grok. Regulators, courts, and rival labs will likely watch closely as more details emerge about how companies train models and where they draw the line between inspiration and copying. For readers and users, this story points to a bigger reality: the next phase of the AI race may hinge less on flashy launches and more on the rules of imitation.