Elon Musk Admits xAI Used OpenAI Models to Train Grok in Court Testimony

[cryptobro]

Elon Musk confirmed under oath in a California federal court on Thursday that xAI used distillation techniques on OpenAI models to help train Grok, the AI assistant at the center of his company's competitive push against established frontier labs. When asked directly whether xAI had used distillation, Musk initially framed it as a general practice across the AI industry before confirming "partly" when pressed on whether that applied to xAI specifically. The admission is notable on multiple levels given that Musk is actively suing OpenAI, CEO Sam Altman, and co-founder Greg Brockman in the same proceeding, alleging they breached OpenAI's original nonprofit mission by transitioning to a for-profit structure. The trial provided an inadvertent window into industry practices that most labs have been careful not to discuss publicly.

Distillation, the process of training new models by systematically prompting publicly accessible chatbots and APIs to extract knowledge from them, has been a flashpoint in the AI industry primarily in the context of Chinese labs using the technique to create capable open-weight models at a fraction of the cost of building from scratch. OpenAI, Anthropic, and Google have reportedly launched a joint initiative through the Frontier Model Forum to share information about combating distillation attempts, focusing on preventing suspicious mass queries. Whether distillation is explicitly illegal remains unclear, but it almost certainly violates the terms of service governing use of these products. Musk's admission that xAI engaged in the practice, even partly, is the most direct confirmation yet that the technique is not confined to Chinese competitors but is used within the US AI industry as well. Later in his testimony, Musk ranked the current AI landscape, placing Anthropic first, followed by OpenAI, Google, and Chinese open source models, while describing xAI as a much smaller operation with just a few hundred employees.

[cryptohog]

Elon Musk sued OpenAI for betraying its mission and then confirmed under oath that xAI used OpenAI's models to train Grok, which is a very specific sequence of events.