Sounding the alarm in AI and National Security

The Framework for Artificial Intelligence Diffusion

TL;DR: New U.S. regulations, announced on January 13, 2025, on the export of advanced artificial intelligence (AI) models and advanced computing integrated circuits (ICs). The regulations aim to prevent these technologies from falling into the hands of adversaries, primarily China, and incentivize the purchase of American products from American companies. Specific controls target model weights of advanced AI models and large IC clusters with caps and license exceptions and review policies depending on the destination and end-user. The political and industry reactions to the announcement are mixed, questioning its timing, form, and content.

In this note, I will review how we got here, summarize the new rule, and share initial reactions to the announcement. You can jump directly to the section you're interested in.

How we got here

It is helpful to review the technical timelines leading to the newly announced Framework for Artificial Intelligence Diffusion to understand where we are. Much has changed over the last year. Starting last spring, some began questioning if Deep Learning and large language models (LLMs), the prevailing technology underpinning Generative AI, were hitting a wall with the end of the "scaling laws" that had fueled the recent boom in AI. As recently as November 2024, the press debated this trend's impact on startup valuations, questioned the economics of AI, and speculated if the "AI bubble" was about to burst.

Then, in December, a new dialogue started about achieving a new technical milestone, bringing us closer to the mythical Artificial General Intelligence (AGI – when machine intelligence can understand or learn any intellectual task that a human being can). Shortly before Christmas, the ARC Prize team – a group of scientists and developers who created an AI benchmark to measure progress towards AGI – announced that the latest OpenAI model had achieved a score of 87.5%. In their words:

“This is a surprising and important step-function increase in AI capabilities, showing novel task adaptation ability never seen before in the GPT-family models. For context, ARC-AGI-1 took 4 years to go from 0% with GPT-3 in 2020 to 5% in 2024 with GPT-4o. All intuition about AI capabilities will need to get updated for o3.”

Sam Altman, CEO of Open AI, mused on this year-end reflections:

“We are now confident we know how to build AGI as we have traditionally understood it. We believe that, in 2025, we may see the first AI agents “join the workforce” and materially change the output of companies. […] We are beginning to turn our aim beyond that, to superintelligence in the true sense of the word. We love our current products, but we are here for the glorious future. With superintelligence, we can do anything else. Superintelligent tools could massively accelerate scientific discovery and innovation well beyond what we are capable of doing on our own, and in turn massively increase abundance and prosperity.”

In a January 6, 2025, Wall Street Journal Op-ed, Dario Amodei (CEO of Anthropic) and Matt Pottinger (former Deputy National Security Advisor of the United States during the first Trump administration) alerted the president-elect to the importance of U.S. leading the world in artificial intelligence to preserve national security and highlighted:

“AI will likely become the most powerful and strategic technology in history. By 2027, AI developed by frontier labs will likely be smarter than Nobel Prize winners across most fields of science and engineering. designing new weapons or curing diseases. […] The nations that are first to build powerful AI systems will gain a strategic advantage over its development. […] Export controls, which ban shipments to China of the high-end chips needed to train advanced AI models, have been a valuable tool in slowing China’s AI development. These controls began during the first Trump term and expanded under the Biden administration to cover a wider range of chips and chip-manufacturing equipment. The controls appear to have been effective. […] China is trying to work around U.S. controls, including by using shell companies to set up data centers in countries that can still import advanced U.S. chips. […] The Trump administration should shut down this avenue of circumvention.”

It is essential to do your research and apply judgment to separate hype from reality and identify where speech is motivated by commercial interests or actual scientific achievement. My take is that the truth is somewhere in the middle. These developments set the stage for the policy initiative, with government officials emphasizing the need to act quickly to preserve America's six- to 18-month advantage in AI over rivals such as China.

What was announced

The Bureau of Industry and Security (BIS) of the U.S. Department of Commerce has issued an interim final rule to enhance export controls on advanced computing integrated circuits (ICs) and impose new controls on AI model weights to protect U.S. national security.

An "interim final rule" (IFR) is a regulatory rule issued by a government agency that becomes effective immediately without first going through the typical notice-and-comment period, meaning the public is not given a chance to provide feedback before the rule takes effect. However, the agency usually solicits comments after the rule is implemented and may modify it based on that feedback. IFRs are used when the agency believes there is a compelling reason to enact the rule quickly, such as in an emergency.

The rule introduces global licensing requirements for exporting advanced computing ICs and AI model weights to manage risks associated with their proliferation. It outlines a framework for responsibly controlling AI diffusion, emphasizing stringent security measures for storing model weights and constructing large IC clusters. Comments on this rule must be submitted by the specified deadline: 120 days after the date of publication in the Federal Register. Since the rule is scheduled to be published on January 15, 2025, the comment deadline will be May 15, 2025.

In simpler terms, the IFR creates three country groups.

The first group includes 18 countries seen as close allies, aligned on technology competition with China, and with a history of trusted cooperation on National Security-sensitive technology. Additionally, they should have historically demonstrated a track record of limiting smuggling. No restrictions apply to these countries. The original announcement includes Australia, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Republic of Korea, Spain, Sweden, Taiwan, and the United Kingdom.

The second group represents the restricted countries. Those are the five where the U.S. has had a pre-existing arms embargo - China (People's Republic of China), Russia, North Korea, Iran, Syria - and, additionally, Macau. Transactions with entities located in or headquartered in these countries will generally be subject to a presumption of denial, and they will be given no access to advanced American AI capabilities.

The third group is the rest of the world. For these countries, the rule imposes a multi-part licensing framework:

1. Global Licensing Requirement: Exports, reexports, and in-country transfers of advanced computing ICs and AI model weights to these destinations require a license with more flexible review policies than restricted countries.

2. Country Allocations: BIS establishes per-country Total Processing Performance (TPP) allocations that limit the cumulative quantity of advanced computing ICs allowed in each destination. Exports beyond these limits will be denied unless specific national security assurances are provided.

3. Validated End-User (VEU) Authorization: Entities in these countries can apply for National Validated End-User (NVEU) status, allowing them to receive larger quantities of advanced ICs under certain security conditions.

4. License Exceptions: For smaller, low-risk transactions (e.g., involving limited quantities of ICs), exporters may use specific license exceptions by meeting certification and reporting requirements.

In addition, the IFR defines Universal Validated End-User (UVEU) status as entities that meet high security and trust standards and are headquartered in close allied countries (the first group). They can receive advanced computing ICs and AI model weights under a streamlined process and undergo security reviews of new data center locations.

Let’s use an example. A country in the rest of the world can buy up to a cap on total processing power equivalent to roughly 50,000 leading NVIDIA GPUs. If the country is willing to sign an MOU with the United States saying that they will decouple from China on AI technology, they can buy an additional 50,000 (up to 100,000 equivalent GPUs). In this example, Brazil (in the third group) could buy up to 50,000 GPUs and sign an agreement with the U.S. to increase the cap to another 50,000. Additionally, if a hyperscale cloud provider like Microsoft applies for UVEU status - which includes additional cyber and physical security measures for their Brazilian operation - they can open a data center in Brazil with additional capacity without affecting the country cap.

This represents a clear incentive to choose American products or products from countries aligned with the U.S. national security and foreign policy goals.

Reactions to the Framework for Artificial Intelligence Diffusion

The reactions to the announcement are mixed. The Information Technology Industry Council (ITI) came out with a statement along these lines:

“ITI has serious concerns regarding this global export controls rule on AI-related advanced compute tools, which threatens to fragment global supply chains and discourage the use of U.S. technology. Rushing a consequential and complex rule to completion without robust stakeholder engagement pays little regard to its impact on U.S. competitiveness and global alliances – which the Biden Administration has worked hard to strengthen. We urge the incoming Trump-Vance Administration to withdraw the rule and engage with industry so that we can work together to craft effective policies to address these national security concerns.”  

NVIDIA strongly opposes the ruling, saying:

“In its last days in office, the Biden Administration seeks to undermine America’s leadership with a 200+ page regulatory morass, drafted in secret and without proper legislative review. This sweeping overreach would impose bureaucratic control over how America’s leading semiconductors, computers, systems and even software are designed and marketed globally. And by attempting to rig market outcomes and stifle competition — the lifeblood of innovation — the Biden Administration’s new rule threatens to squander America’s hard-won technological advantage.”

The statement by Ned Finkle, vice president of government affairs at NVIDIA, speaks directly to the incoming Trump administration. It needs to be taken with perspective since NVIDIA has a direct stake in the discussion. They currently operate in a supply-constrained environment, with the demand for their products far exceeding their production capacity (NVIDIA GPUs are manufactured in Taiwan). When and if supply catches up with demand, they will need new markets, and any trade restrictions are not in their interest.

Other industry players, like Microsoft, provided a more balanced commentary, including this quote from their president, Brad Smith:

"We're confident we can comply fully with this rule's high security standards and meet the technology needs of countries and customers around the world that rely on us."

Earlier this month, in his “The Golden Opportunity for American AI” memo, Mr. Smith addressed the topic:

“While the U.S. government rightly has focused on protecting sensitive AI components in secure datacenters through export controls, an even more important element of this competition will involve a race between the United States and China to spread their respective technologies to other countries. […] Hence, the United States needs a smart international strategy to rapidly support American AI around the world. […] The best response for the United States is not to complain about the competition but to ensure we win the race ahead. This will require that we move quickly and effectively to promote American AI as a superior alternative. And it will need the involvement and support of American allies and friends.”

The international reaction reflects a concern with creating a hierarchy of countries. The EU issued this statement by Executive Vice-President Henna Virkkunen and Commissioner Maroš Šefčovič:

“We are concerned about the US measures adopted today restricting access to advanced AI chip exports for selected EU member states and their companies. We believe it is also in the US economic and security interest that the EU buys advanced AI chips from the US without limitations: we cooperate closely, in particular in the field of security, and represent an economic opportunity for the US, not a security risk.”

What happens next is hard to predict. The rule will take effect in 120 days, and the new administration can decide to reverse it or use it as a negotiating tool, as countries are kept or taken off the different groups. This is akin to the strategy being used with tariffs.

According to Politico, Senate Commerce Chair Ted Cruz (R-TX) says the rule threatens to cripple innovation and “Worse, the Biden [administration] is exploiting an emergency carveout to bypass standard transparency and accountability procedures to try and jam President Trump and his team.”

On the other side Reps. John Moolenaar (R-MI) and Raja Krishnamoorthi (D-IL), chair and ranking member of the House Select Committee on the Chinese Communist Party, support the effort as demonstrated in their letter to the Secretary of Commerce issued before the announcement. The new administration's nominated Secretary of State, Senator Marco Rubio (R-FL), and designated incoming National Security Advisor, Rep. Mike Waltz (R-FL), have historically been very hawkish on China.

Whatever direction this takes, the Biden administration has decided to sound the alarm on AI and National Security. Stay tuned to see what happens next.