Who’s Afraid of Chinese Models?

The United States faces a strategic choice regarding the regulation of artificial intelligence models. Current industry practices often involve labs training their systems on unlicensed data while simultaneously prohibiting others from using their models to train smaller, competing systems through a process called distillation. To address this inconsistency and bolster domestic competitiveness, a proposal suggests that federal law should explicitly classify AI training as fair use and invalidate terms of service that forbid distillation.

This approach would provide legal protection for AI labs while ensuring that the knowledge gained from these models remains accessible for further innovation. By embracing open development, the United States could better compete with international counterparts, particularly as Chinese firms shift their strategies. For instance, Alibaba recently released its powerful Qwen 3.8 Max model with open weights, a move that aligns with recent directives from President Xi Jinping encouraging open-source collaboration and information sharing.

The Qwen 3.8 Max model itself represents a significant technical milestone, featuring 2.4 trillion parameters and sophisticated reasoning capabilities. These models demonstrate an increasing ability to follow complex instructions and refine their outputs through internal logic, as seen in their detailed planning processes. Establishing a clear, open policy framework is essential for the United States to maintain its leadership in the field, as it would foster a more collaborative ecosystem that mirrors the rapid advancements occurring globally.