Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

NewsPolicy

A bipartisan House bill would have NIST write standards against distillation attacks on US models by China, Russia, Iran and North Korea

CyberScoop reported on October 22 that Representatives Pat Harrigan, a North Carolina Republican, and George Whitesides, a California Democrat, plan to introduce the AI Cyber Defense Act. It would direct NIST to define distillation attacks, in which one model is trained on the outputs and reasoning of another, and distinguish them from the sharing between open-weight and open-source models that drives innovation; to develop voluntary standards and best practices for preventing such attacks by entities of concern, meaning people and organizations headquartered in or directed from China, Russia, Iran and North Korea, with the NIST director able to designate others; and to convene an AI Cybersecurity Roundtable of government and industry to write voluntary standards for the defensive use of AI in critical infrastructure. The article recalls that the government, Anthropic, OpenAI and Google have accused Moonshot AI, DeepSeek and Alibaba of systematic campaigns to distill US commercial models through hundreds of burner and resold accounts and false identities, with thousands or millions of prompts. Harrigan said keeping the lead is incredibly expensive; Whitesides said every month adversaries are kept from stealing progress is a month gained to build safeguards.

Why it matters

Distillation is moving from a lab complaint to a policy category with a definition and a standards body behind it. If you expose a model through an API, the controls NIST writes for the labs will become the checklist your customers ask about.

Sources

Read it with the rest of the news in the Learning Center