The Trump administration is divided on how to handle the emergence of Chinese artificial intelligence that now matches the capability of top American systems.
Internal discussions show a split between officials pushing for tighter restrictions and the Commerce Department, which argues those limits cannot be enforced.
The Moonshot challenge
This dispute follows the release last week of the Kimi K3 model by Moonshot AI, a Chinese lab. That release places the system on par with the leading outputs from Anthropic and OpenAI.
Senior White House officials have recently discussed steps to stop Chinese labs from training their own models using American ones. That practice is known as distillation.
Distillation as IP theft
The White House plans to take presidential action after Anthropic claimed last month that Alibaba carried out the largest known distillation attack against it. One source says this action will not come as an executive order.
A White House spokesperson declined to comment on this reporting.
The administration treats distillation attacks with particular seriousness. It announced on Wednesday that Moonshot created Kimi K3 by distilling Anthropic’s Claude Fable 5 model. Treasury secretary Scott Bessent told Fox Business that covert distillation could lead to sanctions, calling it intellectual property theft.
Lutnick has considered ways to encourage top US labs to build their own open weight models to counter China. He has spoken with leaders at several AI labs in recent weeks.
Lutnick appears to take a middle ground on regulation. He imposed export controls on Anthropic to bring them to heel but has been more freewheeling than Cairncross, one source said.
Open weights and security risks
Earlier this month, the biggest AI issue for the administration concerned Anthropic’s Claude Mythos and Fable 5 models. Their ability to find vulnerabilities in government systems and critical infrastructure was the primary concern.
Policy headaches have worsened with the rise of so-called open weight models from China. These models have few or no safeguards to prevent hacking government systems and can be downloaded and used by anyone.
What it means
Creators and developers face a harder reality. Chinese models are now available to anyone without the same safety checks as US systems. That opens the door for unauthorised access to sensitive infrastructure. The government is debating whether to cut off the data flow or try to match the output, but no formal request for input has gone to the Commerce Department yet.
