US-China dispute grows over AI model distillation

US officials accused China’s Moonshot of “industrial-scale distillation” of Anthropic’s AI and threatened sanctions, arguing that large-scale unauthorized use violates terms of service and raises national security concerns. Distillation, a method for learning from existing AI outputs, is described as common practice across the industry and widely used by US companies including Google and xAI.

China’s Commerce Ministry pushed back, accusing Washington of “double standards” and “typical AI hegemony.” The ministry said US AI companies have also distilled Chinese models in research, development, and training, while Google sells a distillation service and Elon Musk’s team used the technique for Grok.

Moonshot says Kimi K3’s gains came from its own architecture, including a new optimizer called Moon Clip, a linear-attention system, and attention residuals. The dispute has widened as US companies now run nearly 60% of their tokens on Chinese models through OpenRouter, while Hugging Face reportedly used a Chinese model in July to respond to an attack that US frontier models could not handle.