A bipartisan pair of House lawmakers is proposing legislation that would put the National Institute of Standards and Technology (NIST) at the center of the U.S. response to AI model theft. The AI Cyber Defense Act, sponsored by Representative Pat Harrigan (R-N.C.) and Representative George Whitesides (D-Calif.), would direct NIST to write voluntary industry standards and best practices for defending AI models against distillation attacks.
Distillation attacks train one AI model on the outputs and reasoning of a more advanced model, letting an attacker replicate proprietary capabilities far more cheaply than developing them. The bill would require NIST to formally define the term and distinguish malicious distillation from legitimate sharing and open-weight model collaboration. It would also let the NIST director designate “entities of concern” — foreign persons or organizations headquartered in Russia, China, Iran, or North Korea, or acting at their direction — and would establish an AI Cybersecurity Roundtable bringing government and industry together to set voluntary standards for defensive uses of AI in cybersecurity, with a focus on critical infrastructure.
One of the bill’s sponsors told CyberScoop that staying ahead of attackers is “incredibly expensive” for U.S. AI companies and that distillation attacks create incentives to steal rather than innovate.
A crowded legislative field
The bill is the latest in a series of overlapping 2026 efforts that treat model theft as both a national security and an economic threat. In April, the House Foreign Affairs Committee advanced the Deterring American AI Model Theft Act (H.R. 8283) by a 43–0 vote, according to Just Security. That bill would direct the executive branch to identify entities conducting model extraction attacks, consider adding them to the Commerce Department’s Entity List, and authorize blocking sanctions, with the Commerce Secretary coordinating with export policy committees, an iaps.ai analysis explains.
Senator Bill Hagerty introduced the Blocking Large-scale Adversarial Distillation Efforts (BLADE) Act (S. 5252) in August, which would publicly expose foreign entities behind distillation campaigns and authorize export controls and Treasury sanctions. In July, Senators Adam Schiff and Jim Banks and Representatives Bob Latta and Whitesides introduced the Collaboration on Adversarial Threats and Security Risks Act, creating an antitrust safe harbor so AI labs and security researchers can share information and coordinate responses — including delaying high-risk deployments — through bodies such as NSA’s AI Security Center and NIST’s Center for AI Standards and Innovation. On September 15, Representatives Mike Lawler and Josh Gottheimer introduced the Stop Rogue AI Act, directing NIST to write standards for discovering, monitoring, and controlling AI agents that federal agencies and contractors would be required to adopt.
Government attribution and guidance
The bills follow a strong shift in official U.S. messaging. A White House memorandum issued April 23 — National Security and Technology Memorandum 4 — stated that foreign entities, principally based in China, are conducting deliberate, industrial-scale campaigns to distill U.S. frontier AI systems, according to a Center for a New American Security report. In September, NSA, CISA, and FBI published a joint advisory accusing China-based AI companies of systematically harvesting outputs from U.S. models to recreate advanced capabilities at a fraction of the development cost. A Lawfare analysis notes the advisory materials name six Chinese laboratories.
What to do
The September NSA–CISA–FBI advisory recommends that AI providers:
- Detect anomalous prompts, accounts, networks, and behaviors, and monitor subscription-to-usage ratios and throughput patterns.
- Subtly alter outputs to suspected distillation attempts to reduce their value.
- Share intelligence across model providers, cloud platforms, and API aggregators.
The advisory also points to NIST’s adversarial machine learning taxonomy (NIST AI 100-2e2025), citing techniques such as differential privacy, pre- and post-training interventions, and prompt instruction formatting as applicable defenses.
Public reporting does not yet give the AI Cyber Defense Act a bill number, introduction date, or committee referral, and it remains unclear how the various proposals’ mandates for NIST, Commerce, Treasury, DOJ, FTC, and CISA would be harmonized if all become law.