US Targets 'AI Distillation'

The Trump administration has launched a critical new strategy to combat "AI distillation," a sophisticated technique primarily used by adversaries like China to replicate advanced American AI models without needing original training data. This pivotal policy shifts focus from hardware export controls to the software stack, with authorities now evaluating restrictions on closed-source model weights and new reporting requirements to safeguard national security against this growing threat.

The Trump administration has launched a new strategy to combat foreign "AI distillation," a technique primarily used by adversaries like China to replicate advanced American AI models without original training data.
This policy marks a significant shift in the US-China AI competition, moving beyond hardware export controls to focus on the software stack, specifically targeting model weights.
AI distillation involves using a powerful "teacher" model to generate synthetic data for a smaller "student" model, allowing for advanced reasoning at a fraction of the original cost and reducing compute burden by up to 100x, as explained by Investing.com.
The Bureau of Industry and Security (BIS) is evaluating restrictions on closed-source model weights and considering new reporting requirements for high-compute frontier models.
According to Dentons, the BIS views the unchecked diffusion of advanced AI model weights as a national security risk, as they could be manipulated and misused by malicious actors.
The White House Office of Science and Technology Policy reported that foreign actors are deploying tens of thousands of proxies and jailbreaking techniques in coordinated campaigns to systematically harvest knowledge from US frontier models.