Elon Musk Admits xAI May Have Used OpenAI's Models in Federal Court Testimony
Key Takeaways
- ▸Musk testified that xAI may have used OpenAI's models via model distillation, a technique that trains smaller, more efficient models by learning from larger ones
- ▸The admission highlights the competitive tension in AI development, as companies once collaborated on safety and validation but now restrict model access
- ▸OpenAI has been investing resources to prevent distillation of its models, particularly targeting Chinese AI labs and international competitors
Summary
During cross-examination in his ongoing legal battle with OpenAI, Elon Musk testified that xAI may have used OpenAI's AI models to train its own systems through model distillation—a technique where a smaller model learns from a larger one. When pressed by OpenAI attorney William Savitt about whether xAI had practiced distillation with OpenAI's models, Musk initially deflected, claiming it's "standard practice" across the AI industry to use other AI systems for validation and improvement. Distillation is a common technique in AI development, but OpenAI has been actively working to prevent competitors from distilling its proprietary models, particularly targeting Chinese labs like DeepSeek.
The testimony reveals tensions in the competitive AI landscape as companies increasingly restrict access to their models. Anthropic has already cut off both OpenAI and xAI from accessing its Claude coding models, citing terms-of-service violations and competitive concerns. The admission comes amid broader government efforts to prevent foreign actors from distilling American AI technology, with the Trump administration and White House office of science and technology policy taking steps to monitor and prevent distillation by Chinese companies. Musk's testimony underscores the friction between collaborative AI development practices and the proprietary interests of major AI companies.
- Major AI companies are increasingly cutting off competitors' access—Anthropic recently blocked both OpenAI and xAI from its Claude models
- The U.S. government is actively working to prevent foreign distillation of American AI models as part of national competitiveness strategy
