Ben Thompson proposes that the United States pass legislation clarifying that collecting data for training models constitutes fair use while simultaneously banning terms of service that forbid distillation. He argues that preventing users from querying APIs to extract model weights is nearly impossible and that current policies create hypocrisy since labs train on unlicensed data yet restrict how others learn from their outputs.
This approach could allow American open-weight models to compete more effectively against Chinese counterparts by removing legal barriers to knowledge transfer. Alibaba recently reversed its decision to withhold Qwen 3.8 Max weights, releasing them as open source after a speech by Xi Jinping encouraged openness and collaboration. The shift suggests that policy changes in Washington might align with strategic moves in Beijing to foster further innovation.
* Legislation would explicitly protect training data collection under fair use principles.
* Terms of service forbidding distillation would be illegal for US companies.
* Alibaba released Qwen 3.8 Max weights following government encouragement of open source sharing.
