House bill would direct NIST guidance on AI distillation attacks
Original title: Exclusive: bipartisan bill would develop federal guidance for distillation attacks, convene industry on defensive cyber
The AI Cyber Defense Act covers entities tied to Russia, China, Iran, and North Korea and would add an industry roundtable.
Reps. Pat Harrigan and George Whitesides are sponsoring the AI Cyber Defense Act, shared with CyberScoop before release. It would direct NIST to write voluntary standards for preventing distillation attacks by foreign entities linked to Russia, China, Iran, and North Korea, after first defining those attacks apart from open-weight sharing. The bill would also create an AI Cybersecurity Roundtable on defensive AI use in critical infrastructure. Harrigan argues adversaries can copy U.S. models at a fraction of the cost; the article recounts prior U.S. accusations against Moonshot AI, DeepSeek, and Alibaba.
Unconfirmed: Allegations that named Chinese firms ran systematic distillation campaigns are accusations cited by the article, not findings of the bill.
- NIST could also designate specific people or organizations as entities of concern.
- Distillation is described as training one model on another model's outputs and reasoning.
- The article says campaigns used burner accounts and large prompt volumes to extract outputs.
- Whitesides framed the bill as slowing theft of U.S. model progress by China and other adversaries.
- Organizations
- NIST
From the original article
4 min read at cyberscoop.com
One of the bill’s sponsors told CyberScoop it’s “incredibly expensive” for U. S. AI companies to stay ahead and distillation attacks incentivize stealing over innovation.
By djohnson · cyberscoop.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.