White House split over how to curb China's AI distillation attacks

The Trump administration is divided over how to respond to China's rapid AI progress, particularly the use of distillation—training AI models on outputs from leading US systems. The White House, backed by security concerns, is pushing for tougher controls after China's Moonshot AI released its Kimi K3 model, which rivals Anthropic's Claude and OpenAI's GPT. Anthropic recently accused Alibaba of launching the 'largest known distillation attack' to date. Treasury Secretary Scott Bessent labeled the practice 'IP theft' and hinted at potential sanctions. The White House is weighing presidential action, though not an executive order, while the Commerce Department argues such restrictions are unworkable.

Commerce Secretary Lutnick has proposed alternative approaches, including incentivizing US labs to release open-weight models that could counterbalance Chinese offerings. The debate has intensified with the rise of China's open-weight AI models, which lack safeguards and can be freely downloaded, raising hacking and security risks. Commerce, which oversees export controls through the Bureau of Industry and Security, appears to be taking a middle-ground stance—imposing some controls but avoiding heavy-handed measures. The policy rift reflects broader tension over balancing innovation and national security in the face of rapidly advancing Chinese AI capabilities.