Framework for Artificial Intelligence Diffusion

<RULE> DEPARTMENT OF COMMERCE <SUBAGY>Bureau of Industry and Security</SUBAGY> <CFR>15 CFR Parts 732, 734, 740, 742, 744, 748, 750, 762, 772, and 774</CFR> <DEPDOC>[Docket No. 250107-0007]</DEPDOC> <RIN>RIN 0694-AJ90</RIN> <SUBJECT>Framework for Artificial Intelligence Diffusion</SUBJECT> <HD SOURCE="HED">AGENCY:</HD> Bureau of Industry and Security, Department of Commerce. <HD SOURCE="HED">ACTION:</HD> Interim final rule; request for comments. <SUM> <HD SOURCE="HED">SUMMARY:</HD> With this interim final rule, the Commerce Department's Bureau of Industry and Security (BIS) revises the Export Administration Regulations' (EAR) controls on advanced computing integrated circuits (ICs) and adds a new control on artificial intelligence (AI) model weights for certain advanced closed-weight dual-use AI models. In conjunction with the expansion of these controls, which BIS has determined are necessary to protect U.S. national security and foreign policy interests, BIS is adding new license exceptions and updating the Data Center Validated End User authorization to facilitate the export, reexport, and transfer (in-country) of advanced computing (ICs) to end users in destinations that do not raise national security or foreign policy concerns. Together, these changes will cultivate secure ecosystems for the responsible diffusion and use of AI and advanced computing ICs. </SUM> <EFFDATE> <HD SOURCE="HED">DATES:</HD> <E T="03">Effective date:</E> This rule is effective January 13. 2025. <E T="03">Compliance date:</E> As explained in the <E T="02">SUPPLEMENTARY INFORMATION</E> section, exporters, reexporters, and transferors are not required to comply with the changes made in this rule until May 15, 2025, except that paragraphs 14, 15, and 18 of supplement no. 10 to part 748 have a delayed compliance date of January 15, 2026. <E T="03">Comment due date:</E> Comments on revisions and additions in this rule are strongly encouraged and must be received by BIS no later than May 15, 2025. <E T="03">Saving clause:</E> Shipments of items removed from eligibility for a License Exception or export, reexport, or transfer (in-country) without a license (NLR) as a result of this regulatory action that were en route aboard a carrier to a port of export, reexport, or transfer (in-country), on May 15, 2025, pursuant to actual orders for export, reexport, or transfer (in-country) to or within a foreign destination, may proceed to that destination under the previous eligibility for a License Exception or export, reexport, or transfer (in-country) without a license (NLR), provided the export, reexport, or transfer (in-country) is completed no later than on June 16, 2025. </EFFDATE> <HD SOURCE="HED">ADDRESSES:</HD> Comments on this rule may be submitted to the Federal rulemaking portal ( <E T="03">www.regulations.gov</E> ). The <E T="03">regulations.gov</E> ID for this rule is: BIS-2025-0001. Please refer to RIN 0694-AJ90 in all comments. All filers using the portal should use the name of the person or entity submitting the comments as the name of their files, in accordance with the instructions below. Anyone submitting business confidential information should clearly identify the business confidential portion at the time of submission, file a statement justifying nondisclosure and referring to the specific legal authority claimed, and provide a non-confidential version of the submission. For comments submitted electronically containing business confidential information, the file name of the business confidential version should begin with the characters “BC.” Any page containing business confidential information must be clearly marked “BUSINESS CONFIDENTIAL” on the top of that page. The corresponding non-confidential version of those comments must be clearly marked “PUBLIC.” The file name of the non-confidential version should begin with the character “P.” Any submissions with file names that do not begin with either a “BC” or a “P” will be assumed to be public and will be made publicly available through <E T="03">https://www.regulations.gov</E> . Commenters submitting business confidential information are encouraged to scan a hard copy of the non-confidential version to create an image of the file, rather than submitting a digital copy with redactions applied, to avoid inadvertent redaction errors which could enable the public to read business confidential information. <FURINF> <HD SOURCE="HED">FOR FURTHER INFORMATION CONTACT:</HD> <E T="03">For general questions contact:</E> Hillary Hess at 202-482-2440 or <E T="03">RPD2@bis.doc.gov</E> . <E T="03">For technical questions contact:</E> <E T="03">Category 2:</E> Sean Ghannadian at 202-482-3429 or <E T="03">Sean.Ghannadian@bis.doc.gov</E> . <E T="03">Category 3:</E> Carlos Monroy at 202-482-3246 or <E T="03">Carlos.Monroy@bis.doc.gov</E> . <E T="03">Category 4:</E> Aaron Amundson at 202-482-0707 or <E T="03">Aaron.Amundson@bis.doc.gov</E> ]. </FURINF> <SUPLINF> <HD SOURCE="HED">SUPPLEMENTARY INFORMATION:</HD> <HD SOURCE="HD1">I. Background</HD> BIS is amending the EAR to enhance and refine its framework for applying export controls to regulate the global diffusion of the most advanced artificial intelligence (AI) models and large clusters of advanced computing integrated circuits (ICs) to protect U.S. national security and foreign policy interests. Specifically, BIS is expanding existing controls on advanced computing ICs controlled under ECCNs 3A090.a and 4A090.a and the corresponding .z items and imposing new controls on the model weights of certain advanced closed-weight dual-use AI models controlled under newly created ECCN 4E091. Background on these changes, which follow extensive U.S. government consideration of the impact of advanced dual-use AI models on U.S. national security and foreign policy interests, is detailed below. The Export Control Reform Act of 2018 (ECRA), 15 U.S.C. 4801, <E T="03">et seq.,</E> emphasizes that national security “requires that the United States maintain its leadership in the science, technology, engineering, and manufacturing sectors, including foundational technology that is essential to innovation.” ECRA also commands that BIS's rules be “transparent, predictable, and timely,” and have “the flexibility to be adapted” in light of evolving circumstances. Given that this rule imposes new global requirements on a rapidly developing industry that is central to innovation, BIS has determined that it can best further the United States's technological leadership, and provide transparency and predictability to affected stakeholders, by delaying the compliance date of this rule, as outlined in the <E T="02">DATES</E> section. Such delay will allow stakeholders time to familiarize themselves with the rule, and to provide comment on it, before compliance is required. <HD SOURCE="HD2">a. Risks and Benefits of Advanced AI Models</HD> Over the past decade, AI models have shown striking performance improvements across many domains, including reasoning, coding, and image and voice recognition and creation. These performance improvements create numerous direct applications, such as text generation and computer code-assistance, and can increasingly support AI agents that are able to interact with digital and physical systems, increasing the ability of laypeople to, for example, write code or use scientific tools that previously required specialized skills. Performance improvements may continue as AI developers increase the scale and efficiency of their models. Experts from across the U.S. government have determined that as the capabilities of these models continue to improve, they will enable malicious actors to engage in activities that pose profound risks to U.S. national security and foreign policy objectives. As the Office of the Director of National Intelligence has assessed, AI models have the potential to enable advanced military and intelligence applications; lower the barriers to entry for non-experts to develop weapons of mass destruction (WMD); support powerful offensive cyber operations; and assist in human rights violations, such as through mass surveillance. <E T="03">See</E> Feb 5, 2024, Office of the Director of National Intelligence, Annual Threat Assessment of the U.S. Intelligence Community. For example, a dual-use AI model trained on data describing the functions and mechanics of chemical compounds or biological sequences could lower barriers to the development of chemical or biological weapons by providing protocols and troubleshooting information that would enable non-experts to design and produce such weapons at low cost. Similarly, the Department of Homeland Security has determined that advancements in AI may lower the barriers to entry for WMD development for both state and non-state actors and thus enhance malicious actors' ability to conduct attacks that threaten U.S. national security. <E T="03">See</E> April 26, 2024, Department of Homeland Security Report on Reducing the Risks at the Intersection of Artificial Intelligence and Chemical, Biological, Radiological, and Nuclear Threats. At the same time, experts from across the U.S. government have determined that, in the hands of validated entities operating under secure conditions, dual-use AI models have the potential to create significant economic and social benefits in the United States and across the globe. As BIS explained in a previous rule, dual-use AI models hold the potential to increase access to healthcare, education, and food and assist with combatting complex problems such as climate change. <E T="03">See</E> Expansion of Validated End User Authorization: Data Center Validated End User Authorization, 89 FR 80,080 (Oct. 2, 2024). For example, under the right conditions, advanced AI models could be an enabler ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Preview showing 10k of 337k characters. Full document text is stored and available for version comparison. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━