Who’s Afraid of Chinese Models?

Simon Willison's Weblog reports on a proposal from tech analyst Ben Thompson that tackles a legal contradiction at the heart of AI development. Thompson suggests the U.S. should pass a law making it explicit that collecting data to train AI models is fair use, and preventing companies from using terms of service to ban model distillation—the practice of extracting knowledge from an AI by querying it repeatedly. This addresses a double standard: labs forbid this technique in their terms of service, yet many trained their models on unlicensed data themselves. The proposal could help American open-source models compete more effectively with their Chinese counterparts. Meanwhile, Alibaba recently reversed course on Qwen, releasing its latest model, Qwen 3.8 Max, as open weights after declining to do so with Qwen 3.7 Max in May—a shift Thompson theorizes may have been influenced by recent remarks from Chinese leader Xi Jinping calling for open-source collaboration and innovation.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton