Who’s Afraid of Chinese Models?

This article examines Ben Thompson's proposal for US legislation to address AI model competition with China. It suggests making data collection for training models fair use and barring terms of service that forbid distillation (querying APIs). The piece discusses how this could help US open models compete with Chinese counterparts like Alibaba's Qwen 3.8 Max, which was released as open weights possibly influenced by Xi Jinping's pro-open-source speech. It also touches on the hypocrisy of labs prohibiting distillation despite training on unlicensed data, and includes technical details about model sizes and reasoning traces.