China's AI Distillation Threatens US Leadership in Open-Weight Models (foreignaffairs.com)

• Chinese AI labs (DeepSeek, Moonshot, MiniMax) use fraudulent accounts to distill US models systematically.
• Distilled models lack safety safeguards, posing security risks (e.g., OpenClaw malware).
• US policy options include export controls, FDPR extension, and supporting domestic open-weight models.

"Chinese AI labs (DeepSeek, Moonshot, MiniMax) are systematically distilling capabilities from US frontier models via fraudulent accounts, violating terms of service. This gives China an edge in open-weight models for local devices, bypassing safety safeguards. The article proposes US policy responses including export controls, extending the Foreign Direct Product Rule, and supporting domestic open-weight alternatives."