Musk admits in court that xAI distilled OpenAI models

On May 2, Elon Musk admitted in a hearing for xAI's lawsuit against OpenAI that xAI had distilled competing large language models, including OpenAI's.

The timing makes this notable—the U.S. just introduced legislation targeting model distillation as grounds for sanctions, and Musk volunteered the admission in court. His defense: "This is standard practice across the AI industry." Google, OpenAI, and Anthropic have all used distillation during training, and some have directly trained models on pirated books and public datasets scraped from the web. By that measure, few Silicon Valley AI companies have clean hands.

The contrast with Chinese AI companies sharpens the picture. Domestic firms have largely taken the open-source route, releasing models freely to the developer community. Some Silicon Valley companies have built products on top of these open-source models while obscuring their technical origins. Musk's testimony drags an industry norm into the open—everyone borrows from everyone else, but some hide it while others admit it outright.

The lawsuit's outcome could reset the rules for the entire AI sector. If the court deems distillation legal, Silicon Valley's mutual borrowing continues unchecked. If it rules otherwise, the industry faces a shakeup. Musk's voluntary disclosure suggests either confidence that the law favors his position, or a strategy to implicate all competitors—if no one's clean, no one can claim moral high ground.