Why This Is Noise: The admission confirms what the industry already assumed. No new product, policy, or legal outcome has changed, the trial continues with no near-term business impact.
Key Takeaways
- Under oath, Elon Musk confirmed xAI used distillation techniques on OpenAI models to help train Grok, calling it a general industry practice
- Distillation involves systematically prompting rival models to replicate their capabilities at a fraction of the cost, and may violate terms of service even if not explicitly illegal
- OpenAI, Anthropic, and Google have launched a joint initiative through the Frontier Model Forum to detect and block distillation attempts, particularly from Chinese firms
- Musk ranked the leading AI providers as Anthropic first, followed by OpenAI, Google, and Chinese open-source models, with xAI ranked lower due to its smaller size
- The admission adds irony to frontier labs’ crackdown on distillation, given their own alleged use of copyrighted data to train models
Elon Musk has confirmed under oath that xAI used distillation techniques on OpenAI models to help train Grok, an admission that puts one of the AI industry’s most prominent figures on record engaging in a practice the same industry is now working to stamp out.
Testifying in a California federal court on Thursday as part of his ongoing lawsuit against OpenAI, Musk was asked whether xAI had used distillation on OpenAI’s models. He described it as a general industry practice, and when pressed on whether that meant “yes,” said: “Partly.”
Distillation involves systematically querying a frontier AI model to understand its inner workings, then using those outputs to train a new, cheaper model that approaches similar capability levels.
The technique is not explicitly illegal but may violate the terms of service companies attach to their products. It directly threatens the investment thesis of frontier labs, whose advantages are built on expensive compute infrastructure that distillation allows competitors to effectively bypass.
It’s no surprise that xAI, which started in 2023 years after OpenAI, would try to learn from the then-leader in the field. — TechCrunch
The admission carries significant irony. OpenAI, Anthropic, and Google have been publicly vocal about distillation as a threat, particularly from Chinese firms using it to produce open-weight models at low cost.
Bloomberg reported in April that the three companies launched a joint initiative through the Frontier Model Forum to share intelligence on distillation attempts and prevent mass querying of their systems. Musk’s testimony now confirms at least one American lab was doing the same to a rival.
Distillation is an unresolved legal grey area. Companies relying on proprietary AI models should be aware that their outputs could be used to train competing systems, and that current terms of service enforcement has limits.
Musk’s testimony also included a ranking of the world’s leading AI providers. He placed Anthropic at the top, followed by OpenAI, Google, and Chinese open-source models.
He characterised xAI as a much smaller company with just a few hundred employees, walking back a claim he made last summer that xAI would soon surpass every company besides Google.
OpenAI did not respond to a request for comment on Musk’s admission. The trial, which centres on Musk’s claim that OpenAI’s shift to a for-profit structure breaches its original nonprofit mission, continues this week.
