New Framework Advances AI Innovation While Protecting U.S. National Security
Washington, D.C. — Today, the Department of Commerce’s Bureau of Industry and Security (BIS) announced controls on advanced computing chips and certain closed artificial intelligence (AI) model weights, alongside new license exceptions and updates to the Data Center Validated End User (VEU) authorization. This new regulation serves key U.S. national security and foreign policy interests and supports the Biden-Harris Administration’s broader strategy to cultivate a secure and trusted technology ecosystem for the responsible use and diffusion of AI.
“This policy will help build a trusted technology ecosystem around the world and allow us to protect against the national security risks associated with AI, while ensuring controls do not stifle innovation or US technological leadership,” said U.S. Secretary of Commerce Gina Raimondo. “Managing these very real national security risks requires taking into account the evolution of AI technology, the capabilities of our adversaries, and the desire of our allies to share in the benefits of this technology. We’ve done that with this rule, and it will help safeguard the most advanced AI technology and help ensure it stays out of the hands of our foreign adversaries, while we continue to broadly share the benefits with partner countries.”
“The United States has a national security responsibility to preserve and extend American AI leadership, and to ensure that American AI can benefit people around the world. Today, we are announcing a rule that ensures frontier AI training infrastructure remains in the United States and closely allied countries, while also facilitating the diffusion of American AI globally,” said National Security Advisor Jake Sullivan. “The rule both provides greater clarity to our international partners and to industry, and counters the serious circumvention and related national security risks posed by countries of concern and malicious actors who may seek to use the advanced American technologies against us.”
“AI has been rapidly progressing over the last decade and will only grow more powerful, resulting in the emergence of highly capable models with significant dual-use applications,” said Under Secretary of Commerce for Industry and Security Alan F. Estevez. “This rule will protect national security and advance U.S. foreign policy by ensuring the responsible diffusion of frontier AI technology across the world.”
“Export controls provide a unique tool to address the quintessential dual-use nature of artificial intelligence,” said Acting Assistant Secretary of Commerce for Export Administration Matthew Borman. “Through today’s actions, we ensure the secure spread of AI capabilities, countering the potential for their use in weapons systems and other military activities contrary to U.S. national security. By doing so, we are creating paths that enable trusted partners to use this advanced technology for the benefit of civil society.”
Over the past decade, AI models have shown striking performance improvements across many domains, giving everyday people increased access to tools that previously required specialized skills. As models continue to improve, this increased access will enable malicious actors to engage in activities that pose profound risks to U.S. national security and foreign policy, including enabling the development of chemical or biological weapons; supporting powerful offensive cyber operations; and further aiding human rights abuses, including mass surveillance.
At the same time, AI has the potential to provide tremendous economic and social benefits to humanity. It is impossible to realize the full potential of those benefits without active participation from allies and partners – including like-minded nations, global firms, and research institutions committed to deploying U.S. technology essential to AI development under safe and secure conditions. The Biden-Harris Administration remains committed to ensuring that humanity can reap these critical benefits.
Today’s announcement seeks to keep advanced AI models out of the hands of malicious actors while also ensuring that secure and responsible foreign entities and destinations will have access to the most advanced U.S. AI models, and to the large clusters of advanced computing integrated circuits (ICs) necessary to train those models. Entities and destinations that are willing to abide by certain safety and security mitigations will receive access to AI models and large IC clusters.
The framework adopts a three-pronged strategy.
First, the rule updates controls for advanced computing chips by requiring authorizations for exports, reexports, and transfers (in-country) involving a broad set of additional countries. However, the rule also includes the following license exceptions and authorizations, which will ensure that commercial transactions that don’t pose national security risks can proceed and the benefits of AI can be broadly shared:
Exceptions for certain allies and partners: New License Exception ArtificialIntelligence Authorization (AIA) allows for the export, reexport, or transfer (in-country)of advanced computing chips, without an authorization, to a set of allies and partners.
Exceptions for supply chains: New License Exception Advanced Compute Manufacturing (ACM) allows for the export, reexport, or transfer (in-country) of advanced computing chips, without an authorization, for the purposes of development, production, and storage of these chips, except to arms-embargoed countries. This license exception builds on the Temporary General License from October 2023 rule to prevent disruption of supply chains.
Low volume exception: New License Exception Low Processing Performance (LPP)allows limited amounts of compute to flow globally, except to arms-embargoed countries.
Update to Data Center Validated End User (VEU) Program: The rule further bifurcates Data Center VEUs into:
Universal VEUs (UVEU): Provides U.S. and certain allied and partner country entities with the opportunity to obtain a single authorization that will allow the company to build data centers around the world without additional authorizations, except in arms-embargoed countries.
National VEUs (NVEU): Provides entities headquartered outside arms-embargoed countries the opportunity to obtain an authorization that will allow the company to build data centers in specified locations without additional authorizations, except in arms-embargoed countries.
When a license is required to export or reexport chips to a certain destination, license applications will be reviewed under a presumption of approval until the total quantity of controlled chips exported or reexported to that country exceeds a specified allocation. After a country reaches its allocation, applications will be reviewed under a policy of denial. Consistent with previously established policy, a presumption of denial remains in place for arms-embargoed countries, regardless of quantity.
Authorized NVEUs will be able to build data centers up to a specified scale in each country. This allocation is separate from, and not impacted by, the host country’s specified country allocation. Likewise, the low-volume orders are not affected by and do not count against country-level allocations. Authorized UVEUs will be required to keep at least 75% of their controlled advanced chips within the United States and certain allied and partner countries, and will be prohibited from installing more than 7% of their controlled chips in any single other country. U.S.-headquartered UVEUs will be required to keep at least 50% of their controlled advancedchips in the United States.
Second, the rule institutes new controls on the model weights of the most advanced closed-weight AI models. These controls will initially apply to the weights of models trained with 10^26 computational operations or more, and authorizations will be required to export, reexport, or transfer (in-country) such weights to a broad set of countries. Additionally, the rule creates a new foreign direct product rule that applies these controls to certain model weights produced abroad using advanced computing chips made with U.S. technology or equipment. As with advanced computing chips, however, this rule includes several license exceptions for model weights:
Exception for deployments by U.S., ally and partner-headquartered entities: New License Exception Artificial Intelligence Authorization (AIA) allows for the export, reexport, or transfer (in-country) of otherwise controlled closed AI model weights, without an authorization, by companies headquartered in the United States and certain allies and partners, except to an arms-embargoed country.
Exception for open models: Models with widely available model weights (i.e., open-weight models) are not subject to controls. Additionally, the model weights of closed models that are less powerful than the most advanced open-weight models, even if they exceed the 10^26 threshold, are not controlled.
Third, BIS will impose security conditions to safeguard the storage of the most advanced models in destinations to protect U.S. national security and to mitigate the risk of diversion for advanced computing chips.
Additional Information
BIS’s actions are taken under the authority of the Export Control Reform Act of 2018 and its implementing regulations, the Export Administration Regulations (EAR).
Under these authorities, BIS possesses a variety of tools to control the export of U.S.-origin and certain foreign-produced commodities, software, and technology, as well as specific activities of U.S. persons, for national security and foreign policy reasons.
For more information, please visit BIS’s website at: https:www.bis.gov.
# # #
