Who’s Afraid of Chinese Models?
Simon Willison · Simon Willison · 2026-07-20
Simon Willison summarizes Ben Thompson's policy proposal that the U.S. should codify AI training data collection as fair use and prohibit terms-of-service restrictions on distillation to reduce hypocrisy and help American open models compete with Chinese counterparts.
Appears in
Extraction
Topics: ai-policydistillationcopyrightopen-weight-modelschinese-ai-models
Claims
- U.S. labs are hypocritical in forbidding distillation of their models via terms of service while themselves training on unlicensed data.
- Ben Thompson proposes a U.S. law making data collection for AI training explicit fair use and barring distillation restrictions for U.S. companies.
- Alibaba's reversal on releasing Qwen 3.8 Max as open weights may have been prompted by Xi Jinping's public speech encouraging open source and AI collaboration.
- Stopping distillation via terms of service is practically impossible since it amounts to policing API queries.
Key quotes
The U.S. should pass a law that (1) makes explicit that collecting data for training models is fair use, and (2) bars terms of service that forbid distillation, for U.S. companies at a minimum. Stopping distillation — which is literally just querying the API — is nearly impossible; the U.S. should go the other way.
We should seize this rare, historic opportunity to encourage open source, openness, collaboration and sharing. [Xi Jinping]