This week the Trump administration clarified a stance in its AI policy: it supports open-weight AI models while accusing some Chinese companies of appropriating U.S. technology. The distinction could enable targeted U.S. measures against Chinese AI firms without broadly restricting the open models used by businesses, researchers and startups.
Why this matters
Administration officials drew a line between legitimate, small-scale or authorized model distillation and covert, industrial-scale distillation that they say can amount to intellectual property theft. That differentiation may form the basis for regulatory or trade actions.
Officials’ statements
Michael Kratsios, chief technology adviser to President Trump, wrote on X that “legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem.” He added that “large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.”
Treasury Secretary Scott Bessent, U.S. Trade Representative Jamieson Greer and State Department under secretary Jacob Helberg expressed similar concerns about IP theft.
Greg Brockman, OpenAI co-founder, told Bloomberg that it is too early to determine whether recent Chinese models were distilled from GPT systems. In a briefing with reporters, Brockman told Axios he had not participated in any administration discussions about banning Chinese open models. Brockman described distillation as “definitely a technical issue.”
Two interpretations of distillation
The administration’s messaging effectively distinguishes two forms of distillation:
- Small-scale or authorized distillation that yields cheaper, specialized models and is treated as a normal part of AI development.
- Industrial-scale distillation conducted via fraudulent accounts, evasive access techniques, or violations of U.S. companies’ terms, which is framed as theft.
The Moonshot / Kimi context
Tension has risen around Moonshot’s Kimi model. Kimi’s lower cost has made it an attractive alternative for some users, prompting concern among U.S. policymakers and certain industry players. The White House appears to be preparing the groundwork for actions against what it views as potential IP theft via distillation.
A joint evaluation by the U.K. Artificial Intelligence Security Institute and the U.S. Center for AI Standards and Innovation found that Kimi K3’s cyber capabilities perform substantially below other frontier models, a finding that undermines claims that the model fully matches leading Western systems.
Potential consequences and next steps
Administration comments indicate a desire to keep open AI development intact while taking a hard line where it believes covert, industrial-scale distillation crosses into IP theft. Scott Bessent posted on X that when PRC firms conduct such distillation attacks, “sanctions and Entity List designations will be on the table.”
Previously, the administration has considered adding Chinese AI labs to the Commerce Department’s Entity List — a move that would cut off U.S. access without a license — as first reported by Axios.
Observers will be watching whether the government pursues sanctions, export controls or Entity List actions based on the distinction between routine distillation and alleged industrial-scale theft, even as industry stakeholders weigh competition, security and openness.
Brief technical note on distillation
Distillation is a common AI technique in which a large “teacher” model’s outputs are used to train a smaller “student” model to reproduce the teacher’s behavior. The smaller model can then run more efficiently and at lower cost while producing similar outputs, without repeating the original full-scale training process.
