Elon Musk confirms that xAI used OpenAI models to train Grok

In a federal courtroom in California on Thursday, Elon Musk testified that his artificial intelligence startup, xAI, used OpenAI models to improve its models.

At issue is model distillation, a common industry practice by which a larger AI model acts as a “teacher” of sorts to transfer knowledge to a smaller AI model, the “student.” Although it is often used legally within companies that use one of their AI models to train another, it is also a practice sometimes used by smaller AI labs to try to get their models to mimic the performance of a larger competitor’s model.

When asked on stage if he knew what a distillation model was, Musk said it was using an artificial intelligence model to train another model. When Musk was asked if xAI had mined OpenAI’s technology, he seemed to dodge the question, saying that “generally all AI companies” do something like this. When asked if yes, he said: “Partly.”

When pressed, Musk said: “Using other AI systems to validate your AI is standard practice.”

The distillation model has been on the rise and has sparked more controversy among AI labs in recent years, since the lines of what is legal — and what violates certain company terms or policies — often fall within a gray area. Companies like OpenAI and Anthropic have accused Chinese companies of skewing their models using OpenAI Stated publicly Its concerns about DeepSeek, and Anthropic specifically named DeepSeek, Moonshot, and MiniMax. Google, too, has Steps taken To try to prevent what it calls “distillation attacks” or “a method of stealing intellectual property that violates Google’s terms of service.”

In private anthropy Blog post “Distillation is a legitimate and widely used training method,” the company wrote. “For example, frontier AI labs routinely distill their own models to create smaller, cheaper versions for their customers. But it is also possible to use distillation for illicit purposes: competitors can use it to gain powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently.”