Elon Musk says xAI used OpenAI models to train Grok

OpenAI and Anthropic have recently intensified efforts to push back against third-party attempts to train new artificial intelligence systems using their publicly available chatbots and APIs, a process commonly referred to as “distillation.”

Much of the discussion around distillation has focused on Chinese firms leveraging these techniques to build open-weight models that closely match the capabilities of U.S.-developed systems while operating at significantly lower costs. At the same time, many in the tech industry have long suspected that American AI labs may also rely on similar methods when competing with one another.

That assumption now appears to have some confirmation. During testimony in a California federal court on Thursday, Elon Musk acknowledged that xAI has used distillation techniques involving OpenAI models to develop its Grok system. When asked directly whether this was the case, Musk initially described such practices as common across the industry and, when pressed further, responded, “Partly.”

Musk’s remarks came as part of his ongoing legal dispute with OpenAI and its leadership, including CEO Sam Altman and co-founder Greg Brockman. The lawsuit alleges that OpenAI deviated from its original nonprofit mission by transitioning toward a for-profit model. The trial began this week and includes testimony from Musk.

The admission draws attention because distillation poses a challenge to major AI companies. By allowing others to replicate high-performing models without the same level of investment in computing infrastructure, the technique can erode the competitive advantage built by leading labs. The situation also highlights a broader tension in the industry, where companies have faced scrutiny over their own data collection practices while attempting to restrict how others use their systems.

Given that xAI was founded in 2023, several years after OpenAI established itself as a leader in the field, it is not unexpected that the company would seek to learn from existing models. While distillation is not explicitly illegal, it may violate the terms of service set by AI providers regarding how their tools can be used.

At the same time, OpenAI, Anthropic, and Google have reportedly collaborated through the Frontier Model Forum to address attempts at distillation, particularly those originating from China. These efforts often involve repeatedly querying AI systems to analyse their behaviour and replicate their capabilities. In response, leading labs are working on methods to detect and limit suspicious large-scale query patterns.

OpenAI had not issued a public response to Musk’s comments at the time of reporting.

Later in his court testimony, Musk was also asked about a statement he made the previous year, suggesting that xAI would soon surpass most competitors, except Google. In response, he ranked leading AI organisations, placing Anthropic at the top, followed by OpenAI, Google, and Chinese open-source models. He described xAI as comparatively small, noting that the company currently has only a few hundred employees.