US legislative proposal on distillation
American laboratories developing large language models frequently prohibit users from performing distillation, the process of deriving a smaller, more deployable model from a larger one. According to Simon Willison, an AI researcher and writer, analyst Ben Thompson has proposed that the United States enact legislation that would explicitly classify data collection for model training as fair use and ban contractual terms that forbid distillation, at least for American companies. Thompson argues that distillation bans are practically unenforceable because they amount to mere API querying, and that legislation should instead support open innovation while protecting research laboratories from legal risks. Willison notes that this approach could indemnify laboratories while ensuring that the resulting insights support further innovation for all.
Chinese open-weight release after political signal
In the Chinese context, speculation has linked a recent policy shift at Alibaba Tongyi Lab, Alibaba’s AI research division, to a speech by President Xi Jinping calling for greater openness, cooperation and sharing. Willison reports that following this speech Alibaba released Qwen 3.8 Max, a 2.4 trillion parameter model, as open-weights. This marks a departure from the decision in May 2026 not to release Qwen 3.7 Max. The new model is nearly as large as the 2.8 trillion parameter Kimi K3 from Moonshot AI and is described as a new 2.4T parameter model. The open-weights release means developers and researchers can now download, modify and use the model for their own applications without licence fees. TLDR AI, a technology newsletter, adds that Qwen 3.8 brings improved coding and content generation capabilities, increasing its appeal for enterprises and developers alike. This openness contrasts with the American proposal, which would enshrine a right to distillation but would not require producers to release open weights.
Apple integration shows international adoption
Apple Intelligence recently received approval from Chinese regulators and integrates Qwen models into its products, according to TechCrunch AI. This partnership underscores that Chinese models are gaining international recognition and are being deployed in commercial ecosystems, which may strengthen their competitiveness against Western closed solutions.
Dual regulatory pressure on global AI market
The two developments create a dual pressure. On one side, legislative proposals in the United States seek to protect innovation by permitting distillation. On the other, Chinese political initiatives promote open weights and broader technology sharing. Both trends can influence the global AI market, where companies will need to consider not only technical performance parameters but also legal frameworks and political contexts.
Implications for companies and auditors
For firms implementing AI, the debate creates a need to monitor not only the technical properties of models but also the legal terms governing their use. In the United States, legislative change could ease auditing and replication of models through distillation, while Chinese open weights may offer an alternative source of technology without licensing restrictions. Auditors should prepare procedures that account for possible changes in copyright policy as well as risks associated with using models under different regulatory regimes.