Elon Musk testifies that xAI trained Grok on OpenAI models

During testimony in a California federal court, Elon Musk acknowledged that his AI company xAI employed distillation techniques on OpenAI's models to help train its Grok chatbot. Distillation involves systematically querying a publicly accessible model (like GPT-4) to extract behavior and knowledge, then using that data to train a cheaper, often open-weight model. Although Musk only said 'partly' in response to a direct question, the admission confirms what many in the tech industry suspected: even major AI labs engage in this practice to avoid falling behind. The irony is thick—while frontier labs like OpenAI and Anthropic have been cracking down on Chinese firms using distillation to replicate their models, Musk's own company did the same to learn from the then-leader. Distillation isn't clearly illegal but may violate terms of service. OpenAI, Anthropic, and Google have been collaborating through the Frontier Model Forum to detect and block systematic mass queries from China, but the cat is out of the bag on domestic use.

Later in his testimony, Musk provided a ranking of the world's leading AI providers, placing Anthropic at the top, followed by OpenAI, Google, and Chinese open-source models. He characterized xAI as a much smaller player with just a few hundred employees, contrasting with the thousands at OpenAI and Google. His comments come amid his lawsuit against OpenAI, Sam Altman, and Greg Brockman, alleging breach of the company's original nonprofit mission. The disclosure underscores the intense competitive pressure driving even high-profile labs to cut corners, risking further erosion of the legal and ethical norms around AI development. As the trial continues, this admission may become a key data point in debates over fair use, model access, and the sustainability of competing solely through raw compute power.