Who’s Afraid of Chinese Models?

2026-07-20 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Fundamental Awareness, quick

Summary

Ben Thompson, in a July 20, 2026 blog post, proposes a new U.S. copyright policy for AI models to address the hypocrisy of labs using unlicensed data while forbidding distillation. His proposal suggests a law that explicitly declares data collection for model training as fair use and bars terms of service preventing distillation for U.S. companies. This aims to help U.S. open models compete with Chinese counterparts by indemnifying labs and fostering innovation. Concurrently, Alibaba's decision to release Qwen 3.8 Max with open weights, a reversal from their May decision regarding Qwen 3.7 Max, is theorized to be influenced by a July 18, 2026 speech by Xi Jinping, who advocated for open source, openness, collaboration, and sharing.

Key takeaway

For Policy Makers evaluating AI regulation, this analysis suggests considering a dual-pronged copyright approach: explicitly define AI training data collection as fair use while simultaneously barring distillation prohibitions for U.S. companies. This framework could foster domestic AI innovation, enhance competition against foreign models like those from China, and align with global calls for open-source collaboration, as exemplified by recent Chinese policy shifts.

Key insights

A new U.S. copyright policy could declare AI training data collection fair use and prohibit model distillation bans.

Principles

• AI training data collection should be fair use.
• Terms of service should not forbid model distillation.
• Open source fosters innovation and collaboration.

Method

The U.S. should pass a law explicitly defining fair use for AI training data and prohibiting distillation bans in terms of service for domestic companies.

Topics

• AI Copyright Policy
• Model Distillation
• Fair Use Doctrine
• Open-Source AI
• Chinese AI Models
• Alibaba Qwen

Best for: CTO, VP of Engineering/Data, Executive, Policy Maker, Director of AI/ML, Consultant

Related on AIssential

• AI Training Data Copyright 2026: IP Risks & EU/US Rules — Global AI model providers face a structural compliance dilemma due to conflicting EU regulatory obligations and US litigation-driven copyright uncertainty.
• Trump administration reportedly builds a slow-motion ban on Chinese AI models through sanctions and soft pressure — Governments can use "FUD" strategies to restrict foreign tech adoption without direct bans.
• Nadella calls out AI labs like OpenAI and Anthropic for banning distillation while training on everyone else's data — AI labs banning distillation while training on public and customer data creates a "reverse information paradox."
• Copyright and AI Policy Needs Precision, Not Panic — AI policy needs nuanced solutions that protect creators without empowering corporate intermediaries or stifling innovation.
• Microsoft, IFS, TSMC: This Week's Top Five Stories in AI — The global AI landscape is rapidly evolving with infrastructure, ethical, and application advancements.
• Preserving the Freedom to Learn in AI — Preserving the "freedom to learn" is crucial for an open and competitive AI ecosystem.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.