The U.S. wants to contain China’s AI. Silicon Valley keeps using it

The U.S. government is concerned about China's rapid advancements in artificial intelligence (AI), particularly in the realm of frontier models. However, recent events have complicated this narrative. Tarun Chhabra, a senior official at Anthropic who previously worked in the Biden administration, warned about model distillation, a technique where a smaller "student" model is trained using outputs from a larger "teacher" model.

Chhabra specifically cited Zhipu, a Chinese AI developer, as one of the companies accused of distilling capabilities from frontier American models. Interestingly, just hours after Chhabra's comments, OpenAI's former CTO Mira Murati revealed that her new company, Thinking Machines, had incorporated Chinese models into its foundation model.

This contradicts the notion that Chinese firms are solely responsible for catching up to American AI capabilities. The open-weight and source model world operates on a principle of building on top of one another's innovations. While American companies like OpenAI and Anthropic are concerned about unauthorized extraction of their technologies, they are also increasingly incorporating advances from Chinese open-weight ecosystems.

This reciprocal flow of knowledge presents a challenge for policymakers, who had previously assumed that innovation primarily flows from the U.S. outward. As AI advances become increasingly global, policymakers may find it more difficult to enforce durable technological barriers. Rather than preventing knowledge from crossing borders, the future of frontier AI may depend on remaining the most attractive place to develop new ideas.