US Treasury targets Chinese AI models over intellectual property fears

The threat of sanctions follows reports that Chinese models, including Moonshot AI’s Kimi K3, are rapidly closing the performance gap with U.S. counterparts like OpenAI and Anthropic. While Washington has previously focused on restricting hardware exports and chip access, this shift toward targeting the software models themselves reflects growing anxiety over the economic impact of foreign competition on domestic capital raising and market share.

Industry experts remain divided on whether the rapid advancement of Chinese AI is driven by theft or genuine innovation. Critics of the administration's stance point to model distillation—the practice of training smaller systems on the output of larger ones—as a standard industry technique rather than a malicious act. Hugging Face CEO Clem Delangue challenged the narrative that distillation is the primary engine of China’s progress, emphasizing that the country’s success stems from high-caliber research teams and a more collaborative development culture. Meanwhile, domestic legal pressures mount for U.S. firms as well; Anthropic recently reached a $1.5 billion settlement over its own use of copyrighted training data, highlighting the precarious legal landscape surrounding the development of frontier models.