China Rejects US Allegations of Unauthorized AI Model Distillation
China's Ministry of Commerce strongly rejected a joint advisory from US security agencies alleging that six Chinese AI firms engaged in unauthorized 'industrial-scale' distillation of US models. The ministry defended model distillation as a routine, globally accepted technical practice used by developers worldwide, including those in the United States. The dispute highlights escalating geopolitical tensions between the US and China over AI technology, compute dominance, and intellectual property. It demonstrates how standard machine learning optimization techniques like model distillation have become focal points in international tech policy and trade disputes. US security agencies NSA, CISA, and FBI advised US companies to adopt defenses against model extraction via distillation. In response, China asserted that US developers also distill open-source Chinese models and accused US firms of using restrictive terms of service to maintain market monopolies.
## BACKGROUND
Knowledge distillation is an established machine learning technique introduced in 2015 to transfer capabilities from a large model (teacher) into a smaller, more efficient model (student). While distillation is widely used across the industry for model compression and efficiency, major AI providers frequently insert terms of service clauses prohibiting competitors from using their API outputs to train competing models.