What does Preamble of Deterring American AI Model Theft Act of 2026 do?

This bill aims to prevent foreign adversaries from threatening U.S. national security. The bill focuses on preventing the extraction of key technical features from closed-source, American-owned artificial intelligence models. The bill includes provisions for 'other purposes' beyond preventing the extraction of AI model features.

What does SHORT TITLE. of Deterring American AI Model Theft Act of 2026 do?

This bill is formally titled Deterring American AI Model Theft Act of 2026.

What does SENSE OF CONGRESS. of Deterring American AI Model Theft Act of 2026 do?

This section expresses the sense of Congress regarding artificial intelligence (AI) models. Congress believes AI models owned by U.S. private sector companies are vital for U.S. economic and national security. Many advanced AI models are 'closed-source,' meaning their technical details are not publicly available. The unauthorized acquisition of closed-source AI model capabilities, such as model weights, is considered a threat to U.S. national security and economic competitiveness. The government should collaborate with private AI model owners to deter and punish model extraction attacks by entities of concern. Model extraction attacks provide foreign adversaries with a shortcut to acquiring advanced AI capabilities. Authorized model training practices consistent with contractual terms are considered legitimate research and distinct from model extraction attacks.

What does DEFINITIONS. of Deterring American AI Model Theft Act of 2026 do?

This section defines key terms used throughout the Act. “Appropriate congressional committees” are defined as the House Committee on Foreign Affairs and the Senate Committee on Banking, Housing, and Urban Affairs. A “closed-source AI model” is defined as an AI model with proprietary key technical information not publicly shared, access governed by contractual agreements, and provided through owner-controlled interfaces. “Country of concern” includes the People’s Republic of China, the Russian Federation, and other countries designated by the Secretary of State. An “entity of concern” is a foreign person or entity located in, or under the direction of, a country of concern, or conducting model extraction attacks. “Fraudulent account network provider” refers to entities that knowingly provide access to closed-source AI models to entities of concern, circumventing access restrictions. “Model extraction attack” is defined as unauthorized attempts to replicate or improve AI models by circumventing access controls or violating usage terms. The term “owner” refers to the entity holding intellectual property rights to a closed-source AI model and is a United States person.

What does ASSESSMENT OF MODEL EXTRACTION ATTACKS AND FRAUDULENT ACCOUNT of Deterring American AI Model Theft Act of 2026 do?

This section mandates an assessment of model extraction attacks and fraudulent account network providers targeting closed-source AI models owned by United States entities. The Secretary of State, in coordination with relevant agencies, must complete this assessment within 180 days of the Act's enactment. The assessment must identify entities and countries involved in model extraction attacks and fraudulent account networks, analyze attack methods, and evaluate the economic and national security consequences. The Secretary of Commerce will consult with AI model owners, companies, academic experts, and industry fora to identify attacker behavior and develop best practices. A report detailing the assessment's findings will be submitted to Congress within 210 days of enactment, with annual updates for three years. The Secretary of Commerce will establish an information-sharing mechanism for AI model owners to confidentially share information about attacks. The Secretary of State will maintain and publish an 'AI Model Extraction Attackers List' on the Department of State's website, protecting confidential information. The Secretary of Commerce must publish a report on best practices for detecting, preventing, and responding to model extraction attacks within 210 days of enactment.

What does DETERRING MODEL EXTRACTION ATTACKS AND FRAUDULENT ACCOUNT of Deterring American AI Model Theft Act of 2026 do?

This section outlines procedures for deterring model extraction attacks and fraudulent account networks. The Under Secretary of Commerce for Industry and Security, in coordination with the End-User Review Committee, must determine within 210 days of the Act's enactment whether entities involved in such attacks or facilitating them should be added to the Entity List. Entities added to the Entity List may be subject to sanctions, including the blocking of transactions in property and interests in property, as determined by the President through the Secretary of State. Exceptions to these sanctions exist for complying with international obligations, providing humanitarian assistance, and conducting authorized intelligence, law enforcement, or national security activities. Violations of this section and related regulations are subject to penalties under the International Emergency Economic Powers Act.

Back to bills

Deterring American AI Model Theft Act of 2026

Last updated: 4/22/2026 · Introduced: 4/15/2026

Author: Bill Huizenga (R-MI)

TL;DR (AI)

This bill requires an assessment of model extraction attacks and fraudulent account networks targeting American-owned closed-source AI models, identifying involved entities and analyzing the consequences.
The bill establishes an 'AI Model Extraction Attackers List' and allows the government to add entities facilitating these attacks to the Entity List, potentially subjecting them to sanctions.
The Secretary of Commerce will develop and publish best practices for detecting, preventing, and responding to model extraction attacks, and create an information-sharing mechanism for AI model owners.

119th Congressin committeeOfficial page

Verified Votes

No verified votes yet.

Community Votes

No community votes yet. Be the first!

Voting as guest · Sign in for verified votes

Other Sections

Includes provisions on preamble, short title., and 4 more.

6 sections

Click "Show details" to explore individual sections. Use the community vote above or visit All Sections to vote per section.

Read official text

Want the full experience?

Create a free account to cast verified votes on each section, get a personalized summary, and track your voting history.

Create Free Account