Deterring American AI Model Theft Act of 2026

Key Points

• The bill targets foreign adversaries, specifically China and Russia, that steal the inner workings of privately owned American AI models. It defines these thefts as "model extraction attacks" and creates a legal framework to identify, punish, and deter them.
• The State Department would maintain a public "AI Model Extraction Attackers List" naming specific individuals and foreign entities caught conducting these attacks, published on a government website for up to five years.
• The bill also goes after "fraudulent account network providers" -- foreign businesses that sell fake accounts or help adversaries bypass location blocks so they can access American AI systems they are not supposed to use.
• Foreign entities caught stealing AI technology could face sanctions under the International Emergency Economic Powers Act, including having their U.S. assets frozen. They could also be added to the Commerce Department's Entity List, which bans American companies from selling them goods or technology.
• The Commerce Department would set up a confidential information-sharing system so AI companies can quickly report attacks. The government would also publish public best practices for detecting and preventing these thefts within 210 days of enactment.
• The bill carefully protects legitimate AI research, making clear that model training done within the terms of service or with the owner's permission is not considered theft, even if it involves learning from another model's outputs.