Elon Musk testifies that xAI trained Grok on OpenAI model

OpenAI and Anthropic have recently been waging war against third-party efforts to train new AI models by encouraging publicly accessible chatbots and APIs (a process known as “distillation”).

The conversation focuses on how Chinese companies are using distillation to develop open-weight models that perform nearly as well as U.S. products, but are available at a much lower cost. But technology officials widely believe that U.S. laboratories are leveraging these technologies against each other to keep up with competitors.

Now we know that to be true in at least one case. On the stand in a California federal court on Thursday, Elon Musk was asked whether xAI used distillation techniques in its OpenAI models to train Grok, claiming it is a common practice among AI companies. Asked if that meant “yes,” he said “partially.”

Musk is currently suing OpenAI and CEOs Sam Altman and Greg Brockman, accusing them of violating OpenAI’s original nonprofit mission by turning it into a for-profit organization. The trial began this week with testimony from technology leaders.

Musk’s admission is notable because distillation threatens to undermine the advantages that big AI companies have built through investments in computing infrastructure. This allows other software manufacturers to create models with roughly equivalent functionality at a lower cost. There is considerable irony here, given that Frontier Labs has been accused of bending and violating copyright rules in its search for enough data to train its models.

It’s no wonder that Musk’s xAI, which launched in 2023 a few years after OpenAI, would seek to learn from the then-leader in the field. It’s not clear whether distillation is explicitly illegal, but rather it may be a violation of the terms of service that companies have set for users of their products.

OpenAI, Anthropic, and Google have reportedly begun an effort to share information on how to counter distillation attempts from China through the Frontier Model Forum. These typically involve systematic queries to understand the inner workings of the model. To thwart this effort, Frontier Lab is working to prevent users from making suspicious bulk queries.

OpenAI did not respond to a request for comment on Musk’s admission at press time.

Later in his testimony, Musk was asked about claims he made last summer that xAI would soon surpass that of any company other than Google. In response, he ranked the world’s leading AI providers and said Anthropic took the top spot, followed by OpenAI, Google, and China’s open source models. He characterized xAI as a much smaller company with only a few hundred employees.