HR8283-119

What This Bill Does

The Deterring American AI Model Theft Act of 2026 responds to unauthorized extraction of closed-source artificial intelligence models owned by U.S. companies. It defines closed-source AI models by proprietary model weights or architecture, owner-controlled access through APIs or similar interfaces, and contractual limits that prevent third parties from recreating or hosting the model. It defines countries of concern to include the People's Republic of China, Hong Kong, Macau, Russia, and certain Country Group D:5 countries designated by the Secretary of State, and defines entities of concern to include foreign persons or entities linked to those countries or attempting model extraction attacks. Within 180 days, the Secretary of State, coordinating with Operating Committee for Export Policy agencies, must assess entities conducting model extraction attacks and fraudulent account network providers. The assessment must identify countries of origin, government assistance, attack methods, provider office and data-center locations when possible, attack counts from the prior two calendar years, detection approaches, economic and national-security consequences, and U.S. government assistance to model owners. Within 210 days, the Under Secretary of Commerce for Industry and Security and End-User Review Committee agencies must decide by majority vote whether identified attackers, facilitators, or 50-percent affiliates should be added to the BIS Entity List. The President, through the State Secretary, may block property and transactions under IEEPA for identified entities, subject to exceptions for United Nations obligations, humanitarian trade and assistance, and authorized intelligence, law-enforcement, or national-security activities.

Who Benefits and How

U.S. closed-source AI model owners benefit because federal agencies must identify, deter, punish, and assist against model extraction attacks. AI safety and security teams benefit from federal analysis of detection methods, fraudulent account networks, and attack origins. BIS Entity List staff benefit from a statutory process for considering model-theft actors and affiliates. U.S. national-security agencies benefit from an assessment of how model theft affects military and foreign-policy interests. Congressional export-control committees benefit from clearer definitions and later oversight of Entity List and sanctions actions. U.S. AI companies benefit if sanctions and export controls reduce foreign adversary access to proprietary model capabilities.

Who Bears the Burden and How

Entities of concern that conduct model extraction attacks face possible Entity List restrictions and IEEPA blocking sanctions. Fraudulent account network providers face identification, sanctions exposure, and export-control consequences. PRC and Russian AI-linked firms face increased scrutiny when tied to attacks or fraudulent account networks. The Secretary of State must complete the assessment and coordinate with export-policy agencies. BIS and End-User Review Committee members must vote on Entity List treatment within 210 days. Humanitarian and medical trade actors must navigate exceptions when sanctioned entities are involved.

Key Provisions

• Defines closed-source AI models, model extraction attacks, fraudulent account network providers, countries of concern, and entities of concern.
• Requires a 180-day State-led interagency assessment of AI model extraction attacks and fraudulent account network providers.
• Requires the assessment to identify attack origins, foreign government assistance, methods, detection approaches, consequences, and U.S. support to victims.
• Requires BIS and End-User Review Committee agencies to decide within 210 days whether identified entities and affiliates should be added to the Entity List.
• Authorizes IEEPA blocking sanctions for identified entities of concern.
• Provides exceptions for United Nations obligations, humanitarian trade and assistance, and authorized intelligence, law-enforcement, or national-security activities.