U.S. House H.R.8283 "Deterring American AI Model Theft Act of 2026" — treating "model extraction attacks" on closed-source AI as a security threat
A House bill that frames the unauthorized extraction of weights and architecture from U.S.-owned closed-source AI models by foreign adversaries ("model extraction attacks") as a national-security threat, calling to identify, punish, and deter it. Reported out of the House Foreign Affairs Committee 43–0.
Bill overview (primary data)
- Bill numberH.R. 8283
- TypeHouse Bill
- Congress119th Congress
- Latest actionOrdered to be Reported (Amended) by the Yeas and Nays: 43 - 0.（2026-04-22）
Key points
- Frames extraction of closed-source AI model weights/architecture by adversaries as a national-security threat
- Calls for the government and private owners to identify, punish, and deter "model extraction attacks"
- Distinguishes legitimate model training (per terms of service) from extraction attacks
- Reported out of the House Foreign Affairs Committee 43–0 (14 cosponsors as of 2026-06-06)
- Introduced by Rep. Huizenga (R-MI) and others
- Model extraction attacks are prohibited while training consistent with terms of service is expressly stated to be different.
1The concern over AI model theft
H.R.8283, the "Deterring American AI Model Theft Act of 2026," was introduced by Rep. Bill Huizenga (R-MI-4) with Rep. John Moolenaar and others on April 15, 2026. Referred to the House Committee on Foreign Affairs, it was ordered to be reported (amended) by a 43–0 vote on April 22, 2026, with 14 cosponsors as of 2026-06-06.
2The Sense of Congress findings
The bill's Sense of Congress states that: (1) AI models held by U.S. private-sector companies are essential to U.S. economic and national-security interests; (2) many of the most advanced are "closed-source" models whose weights and architecture are not openly shared; (3) unauthorized acquisition of those capabilities by entities of concern through model extraction attacks is a threat to national security, foreign policy, intellectual property, and U.S. competitiveness; and (4) the government, with private owners, should identify, punish, and deter such attacks.
It also distinguishes authorized model training consistent with terms of service from extraction attacks.
3Defining a closed-source AI model
Definitions cover "closed-source AI model" and the appropriate committees (House Foreign Affairs; Senate Banking).
4Drawing the line between attack and research in the text
What marks this bill is that it states not only what it prohibits but what it does not. Set side by side, the line becomes visible.
The opening Sense of Congress sets out that AI models held by US private firms are indispensable to economic and national security, that most frontier models are closed-source with weights and architecture unpublished, and that government should work with private model holders to deter extraction attacks. The bill was ordered to be reported from the House Foreign Affairs Committee, as amended, by a unanimous 43 to 0.
Why it matters
Signals the U.S. placing AI within an economic-security and export-control frame. For firms partnering with leading U.S. AI/semiconductor companies, the framing of model weights/architecture as protected strategic assets — and the line versus legitimate research — is a useful reference for contracts and compliance.
FAQ
What is a "model extraction attack"?
Would it restrict legitimate AI research?
Has it become law?
Sources (primary)
Source: Congress.gov (Library of Congress; U.S. legislative materials, public domain). Links go to the official site.
- Congress.gov (bill page, original)
- H.R. 8283（119th Congress）