Federal Register: Framework for AI Diffusion

government

federalregister.gov·federalregister.gov/documents/2025/01/15/2025-00636/frame...

This January 2025 BIS rule is a landmark US government regulatory action treating frontier AI compute and model weights as strategic national security assets, directly relevant to debates around AI governance, international coordination, and compute-based safety interventions.

Metadata

Importance: 72/100regulationprimary source

Summary

The Bureau of Industry and Security (BIS) establishes a tiered export control framework for advanced AI model weights and computing integrated circuits, dividing countries into three tiers based on trust and national security considerations. The rule aims to prevent adversarial actors from accessing frontier AI capabilities while allowing responsible global AI development among allied nations.

Key Points

•Introduces export controls on advanced AI model weights and high-end AI chips, using a three-tier country classification system based on national security trust levels.
•Tier 1 (close allies) face minimal restrictions; Tier 2 (most countries) face caps on compute imports; Tier 3 (adversarial nations like China/Russia) face near-total restrictions.
•Model weights above a compute training threshold (e.g., ~10^26 FLOPs) are subject to licensing requirements for export to non-Tier-1 countries.
•Establishes verified end-user and data center security requirements for cloud providers and foreign entities seeking to access controlled AI infrastructure.
•Represents a significant escalation of US government involvement in AI governance, directly linking compute access to geopolitical and national security policy.

Review

This Federal Register document represents a significant policy intervention in the global AI technology landscape. The rule establishes a multi-layered approach to controlling the export of advanced AI technologies, focusing specifically on model weights and large computing clusters that could pose national security risks. The methodology involves creating worldwide license requirements, implementing strategic exceptions for low-risk destinations, and establishing detailed security conditions for AI technology transfers. The approach is notably nuanced, seeking to prevent malicious actors from accessing frontier AI capabilities while simultaneously preserving opportunities for responsible international AI development. By creating a graduated control system that considers compute power, destination risks, and end-user validation, BIS demonstrates a sophisticated understanding of the complex technological and geopolitical challenges surrounding advanced AI diffusion.

Cited by 3 pages

Page	Type	Quality
Should We Pause AI Development?	Crux	47.0
Intervention Timing Windows	Analysis	72.0
Hardware-Enabled Governance	Approach	70.0

Cached Content Preview

HTTP 200Fetched Apr 9, 20261 KB

Federal Register :: Request Access 

 
 
 
 
 
 

 
 

 
 

 
 
 
 
 
 
 
 
 

 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 Request Access

 
 

 

 
 Due to aggressive automated scraping of FederalRegister.gov and eCFR.gov,
 programmatic access to these sites is limited to access to our extensive
 developer APIs. Please visit
 
 FederalRegister.gov API
 
 documentation or
 
 eCFR.gov API
 
 documentation to learn more about how to access the API.

 
 Your request has been flagged as potentially automated. If you are human user
 receiving this message, please complete the CAPTCHA (bot test) below and
 click "Request Access". You may occassionally be asked to complete the
 CAPTCHA again, this is normal and part of our security measures.

 unblock#handleSubmit">
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 Request Access
 
 

 
 An official website of the United States government. 

 
 If you experiencing issues with the CAPTCHA or want to request a wider IP range,
 you can use the "Site Help" button found in the lower, right of this page
 to make a request.

 
 
 

 
 
 
 
 
 
 
 
 × 
 
 IP Access Help

 

 
 
 
 This contact form is only for IP Access help.
 Please do not provide confidential information or personal data.
 

 

 * Your Name 

 * Email 

 * How can we help you? 
 

 
 
 

 Upload Attachment 

 * I am requesting technical help. 

 
 
 
 

 
 
 
 
 
 
 
 

 
 

 site-feedback#openModal"
 aria-label="Open site help form">

 
 Site Help

Resource ID: 8e077efb75c0d69a | Stable ID: NzYxNmRkOD