Elon Musk confirms xAI used OpenAI’s models to train Grok

During the ongoing Musk vs. Altman court battle in California, Elon Musk took the stand and admitted that his AI company xAI used OpenAI's models to improve its own Grok models through a technique called model distillation. Distillation is a process where a larger, more capable AI model acts as a 'teacher' to pass knowledge to a smaller 'student' model, making training faster and cheaper. When asked directly whether xAI had distilled OpenAI's technology, Musk initially avoided the question, saying 'generally all the AI companies' do it. Pressed for a clear answer, he replied 'Partly' and later added, 'It is standard practice to use other AIs to validate your AI.'

The admission comes amid rising tensions over model distillation in the AI industry. While the technique is widely used legitimately (e.g., frontier labs create smaller, cheaper versions of their own models), it can also be used illicitly by competitors to replicate powerful capabilities without investing equivalent resources. OpenAI and Anthropic have publicly accused Chinese firms like DeepSeek, Moonshot, and MiniMax of illegally distilling their models. Google has also taken steps to prevent 'distillation attacks,' calling them a form of intellectual property theft. Musk's testimony highlights the gray area of distillation legality, especially when terms of service prohibit using a model's outputs for training competing AI.