119-HR-8283 Journalist Public Summary

Headline Summary

A House bill aims to deter foreign adversaries from copying the secret internals of U.S. company AI models by investigating attacks, naming alleged offenders, coordinating with industry, and enabling trade restrictions and sanctions.

What It Does

Purpose in plain English: stop foreign governments and affiliated actors from stealing how U.S. closed-source AI models work (for example, their hidden “weights”) and punish those who try. It distinguishes allowed research that follows a model’s terms of service from prohibited “model extraction attacks” that use fake accounts, violate usage rules, or evade location and identity checks.

• Directs State and Commerce to assess, within set deadlines, which foreign entities are trying to copy closed models or run networks of fake accounts that help others do so.
• Requires Commerce to consult with AI companies, researchers, and industry groups, share patterns of attacker behavior, and publish public best practices for detection and defense.
• Creates a public “AI Model Extraction Attackers List” naming people and entities found to have conducted such attacks (published for up to five years, with protections for confidential company data).
• Sets up an information-sharing channel so U.S. model owners can confidentially report attacks to the government.
• Triggers consequences: potential addition of identified entities to the Commerce Department’s Entity List (restricting U.S. exports to them) and authorizes the President to block their property and transactions under existing sanctions law, with humanitarian and UN-commitment exceptions.
• Defines key terms (closed-source model, model extraction attack, fraudulent account network provider, entities and countries of concern).

Who’s For It

• Sponsors: Reps. Bill Huizenga (R-MI) and John Moolenaar (R-MI).
• Likely supporters: lawmakers focused on national security and intellectual property protection; U.S. firms that maintain closed, proprietary AI models and want clearer tools to deter abuse.
• What they say: stealing model internals gives adversaries a shortcut to advanced AI, threatens U.S. economic and security interests, and merits export-control-style tools and sanctions.

Who’s Against It

• Potential opponents: civil liberties and open-research advocates worried about chilling legitimate AI research or overbroad interpretations of “extraction.”
• Academic and open-source AI communities may raise concerns about false positives from aggressive detection methods and the impact on interoperability or benchmarking.
• Some business and compliance groups could flag due‑process, evidence standards, and the risks of public naming without robust challenge procedures, as well as extra compliance costs.

What’s Next

Status as of April 16, 2026: Introduced on April 15, 2026, and referred to the House Committee on Foreign Affairs. It awaits committee consideration (hearings, markups) before any House floor vote. If it passes the House, it would move to the Senate.

Key Numbers in the Bill

• Entity List decisions: Commerce’s End-User Review Committee acts by majority vote.
• Sanctions: Authorized under the International Emergency Economic Powers Act, with exceptions for UN obligations and humanitarian activities.

Key Trade-offs and Questions

• Definition precision: Will the criteria for inferring extraction (volume, patterns, coordination) be specific enough to avoid sweeping in benign high‑volume use?
• Research carve‑outs: The bill recognizes authorized research, but implementation details (e.g., honoring contractual permissions, handling TOS ambiguities) will be crucial.
• Attribution challenges: Governments will need strong technical and legal thresholds before publicly naming actors, to reduce misidentification risks.
• International alignment: Success may hinge on partner countries cooperating on export controls and sanctions so listed entities can’t easily route around U.S. measures.
• Industry burden: New reporting and monitoring could improve security but add compliance costs, especially for startups.