xAI & OpenAI · 2026-04-30 · major
Musk Testifies xAI 'Partly' Used Distillation on OpenAI Models to Train Grok
Under oath in his lawsuit against OpenAI, Elon Musk acknowledged xAI 'partly' used distillation on OpenAI models when training Grok, calling it a 'general practice among AI companies.'
On the witness stand in his OpenAI lawsuit, Musk admits xAI 'partly' trained Grok on OpenAI's models — a rare on-record confirmation of cross-lab distillation.
What is it?
Elon Musk took the stand on April 30 in his California federal suit against OpenAI, Sam Altman, and Greg Brockman, which alleges OpenAI breached its original nonprofit mission by going for-profit. Asked directly whether xAI used distillation on OpenAI models to train Grok, he answered 'partly,' and described the technique as a 'general practice among AI companies.'
How does it work?
Distillation here means querying a competitor's hosted model through its public chatbot or API, capturing the outputs, and using them as supervised training signal for a new model. It bypasses the cost of generating fresh teacher data but typically violates the host's terms of service. Musk also volunteered an industry pecking order: Anthropic on top, then OpenAI, Google, then Chinese open-source labs.
Why does it matter?
First time a frontier-lab CEO has acknowledged distilling a US rival on the record. Strengthens OpenAI's pending TOS-violation arguments, hands Anthropic a marketing line, and complicates the regulatory narrative that distillation is mainly a Chinese-lab issue. Expect TOS-clause and output-licensing terms to tighten across all major API providers.