RIN 0694-AJ90
BY JAMES A. LEWIS
May 2025
BIS has amended the EAR to expand its export controls in an attempt to regulate the 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. This EAR amendment is ambitious, but ultimately ineffective at managing the diffusion of AI or preserving any US advantage. Counterintuitively, the rule will actually damage the US global lead in AI and as a consequence, US foreign policy interests and national security.
The issue is not whether exports should be restricted for China. There is little debate over the need for this (although the DeepSeek experience suggests that export controls are overrated when it comes to hampering China’s development of AI). If anything, export controls on China should be expanded to include licensing requirements for China’s ability to access advanced American AI services through the cloud. The only major question is to what extent the availability of equivalent products and services (as defined in BIS regulations) from Chinese sources should be taken into account, since a failure to do so would unnecessarily harm US leadership.
The real issues with this interim rule relate to (a) its erratic application to other countries, and (b) its effect on US technological leadership and national security. The two topics are of course related but can be considered separately, but to avoid further harm to U.S. national security and foreign policy interests, the rule needs significant revision. An administration committed to removing unnecessary regulatory impediments should not accept this inheritance from its predecessors.
The rule’s problems have little to do with the risks or benefits of AI. The risks and potential harms ascribed to AI in the regulation may be generally accurate, but the assumption that export controls reduce these risks is open to question. AI can provide both military and economic advantages if it is applied properly and it may pose a threat to US interests, but this regulation does not reduce that threat. In fact, it unintentionally increases risk to the US by harming the global position of US companies. Sophisticated foreign military and intelligence services will use some form of advanced AI and they will do so regardless of US regulation. While the previous Administration likely overestimated both the benefits and risks of AI, influenced by the larger AI “bubble,” this comment discusses how export controls on AI can best advance the national interest.
Precedents
The Export Control Reform Act of 2018 (ECRA) states that national security “requires that the United States maintain its leadership in the science, technology, engineering, and manufacturing sectors, including foundational technology.” Unfortunately, the practice of export control since the end of the Cold War provides a number of striking examples where a mechanical and heavy handed application of export controls has produced the opposite effect, usually as a result of underestimating both the damage to American industry from ill-advised regulation and the ability of other nations to develop substitute or replacement goods to circumvent US controls. AI is no different. It is worth listing salient example of export control limitations to set the context for a reconsideration of the AI diffusion regulations. This is a pattern that stretches over decades and it is troubling that it must be told again:
- In the 1980s, machine tools exported to the Soviet Union let it make quieter propellors for nuclear submarines. The United States tightened export controls on machine tools, with the result that American companies were driven out of business, leaving the United States dependent on foreign suppliers, including China.
- Until 1997, the U.S. allowed China to launch America-made satellites. The Chinese used this to improve their space launch capabilities (the US gained advantages as well from the arrangement). Concerned over Chinese access, the US blocked all exports of American satellites and their components to China, including European-made satellites with US components and imposed onerous licensing conditions on European satellite makers. In response, the EU created its own component makers (some using subsidies from the European Union) to replace U.S. suppliers. U.S. companies permanently lost market share and smaller U.S. component suppliers exited the satellite business.
- The U.S. controlled and restricted encryption exports until 1999. The result was that despite an initial lead, U.S. software makers were being displaced by foreign software companies until the export restrictions were removed. After that, US firms outcompeted foreign suppliers, leading to a dominant position in internet software and services.
- The United States tried repeatedly to restrict Chinese access to high performance computer chips (CPUs). The famous 1999 commercial of a “dangerous” Mac computer being surrounded by tanks dates from this period. The U.S. had hoped to keep China two generations behind, but chip performance increased rapidly, making control thresholds obsolete, and older chips could be substituted for the most advanced ones to create militarily useful applications. The same is true today. One result was to inspire China to invest more to build its own chip industry.
- The US liberalized rules for the export of commercial imagery satellites but then took an overly cautious approach in approving any sales, thus failing to capture the global market. The advent of “small-sats” allowed foreign producers to build competing satellites that provided adequate (if not “exquisite”) imagery. The US found it preferable to sell imagery as a service, which provided more control over access, rather than imagery satellites themselves. US export rules and risk averse implementation helped speed the growth of foreign capabilities and cost the US markets it could have dominated.
- The US denied Turkish requests to buy American UAVs. In response, Turkey created its own drones (such as Bayraktar), that are cheaper, faster to build, and for many observers, better than US-made drones. The result of denying sales of US drones to global markets is that Turkey, China, and Ukraine now dominate the market.
These were important technologies with strategic significance where the US sacrificed opportunities for global lead as a result of ill-conceived regulation and without affecting opponent capabilities. In the cases of both satellites and encryption, foreign manufactures even advertised their products as “ITAR-free,” meaning not subject to US export controls, successfully attracting foreign customers. Similar outcomes are likely from the AI diffusion rule. An overly risk-adverse approach not attuned to the global dispersion of technology and the increasing difficulty of denying access to it only damages the US advantage and build Chinese competitors.
On of the chief lessons from these experiences are that the US’s has much less ability now, after thirty years of global technology diffusion, to block access to significant dual-use technologies in ways that deny meaningful improvements to opponents’ military capabilities. There have been a few successes, such as designs for military jet engine components, but these have focused on the ‘know how” needed to build them as much as end products or services.” Export controls can make opponent programs more expensive and their products less capable, but not to a degree that will provide advantage to the US or compensate for the damage these rules can do.
It is worth noting that the 1990s efforts to block access to high performance computers and chips were also based on a performance threshold (called MTOPS). This threshold experienced difficulties that ultimately led to it being abandoned. First, the controls could be circumvented by aggregating chips with lower performance thresholds not controlled for export. Second, the steady rate of improvement in chip performance meant that the controls had to be constantly updated to match foreign availability. US export regulations further incentivized China’s efforts to build its own chip industry. This precedent, like the others, is troubling since the likely outcome of this IFR if it is not withdrawn is a stronger Chinese AI and chip industry
One crucial difference between now and the 1980s “golden age” of export controls is the diffusion of knowledge. Part of this diffusion reflects the “open source” technology model, where intellectual property and software are made freely available. For several reasons, including the incentives created by export controls to find non-US AI alternatives, innovators in China and other countries rely on “open” technologies such as open source software, RISC-V chip architectures, and open weight models for AI. Some analysts have concluded that open-weight models are now “sufficiently powerful to underpin a vast array of applications.” DeepSeek, for example, uses an open weight AI model. Meta’s Llama and DeepSeek’s open models provide “access to near state-of-the-art capabilities.” Open technologies are not easily constrained by export controls. A “good enough” AI ecosystem can be created outside the reach of US regulation, thus denying any strategic benefit from export controls. An effort by the US to extend regulatory capture to open source technologies would be quixotic, and access to open source technologies undercuts the effort to manage allocations that undergirds this regulation.
Similarly, a new IFR needs to take into account techniques like quantization. Quantization is an openly available mathematical technique that can be used to reduce an AI model’s memory requirements, accelerates computation, and lower power consumption. DeepSeek uses quantization. While quantization can introduce a small loss of accuracy, careful application of quantization techniques can provide acceptable performance, enabling efficient and practical deployment of AI models without relying on advanced chips or giant data centers. Open source and quantization undercut the IFR’s intended goal of using export controls to creates trusted AI ecosystem managed by the US, particularly in light of continued concern in many countries to protect their “data sovereignty.” This sovereignty concern increases the chances that the IFR will drive other countries to buy from Chinese suppliers.
Regulatory Overreach and Global Industrial Policy
One goal of this regulatory action was to create a “trusted technology ecosystem” managed by the US, to prevent foreign adversaries from gaining access to advanced AI technologies. Creating a “trusted ecosystem” for AI that would be orchestrated out of Washington is industrial policy at a global scale. This is not really even industrial policy, since the intent is not to strengthen US industry but to weaken opponents.
China has in the past “piggybacked” on US technology in its efforts to catch up, but efforts to prevent this must be designed not to harm the US economy or innovation capabilities. These interim final rules were part of a broader strategy intended to prevent the misuse of sensitive technologies by adversarial states, including an ambitious effort to manage the global AI ecosystem, using access to advanced chips to obtain leverage. Unfortunately, in the effort to avoid potential misuse, the IFR damages safeguarding US technological leadership. Technology denial does not provide leadership. A policy whose goal is AI leadership for the US will be discussed later, but it goes without saying that such a pro-growth policy would not include this interim final rule. The two chief objections to it are
— the 168-page rule is complex, cumbersome, and substitutes government decision-making for market-driven outcomes in an attempt to manage the use of AI around the world. GPUs are not a chokepoint in this ecosystem, completely undercutting these ambitions.
— its country groups do not correlate with national security risk. Tier 2 in particular needs to be eliminated by merging it with Tier 1. If a country is hostile to the US and has military or intelligence programs that can take advantage of AI, exports should be denied, but these countries are already defined in Tier 3.
Country Groups
AI is part of a larger transformation driven by computing power and connectivity, which includes fiber-optic networks, satellites, and mobile telecommunications, which provide the foundational infrastructure for future economic growth. Unlike traditional infrastructure, digital networks involve more than physical components—they control and manage intangible assets, services, and data, which are now the primary source of economic growth. Digitalization is reshaping national power in the 21st century, much like industrialization did in the 19th century. The US is well positioned to lead in AI and digitization if it does not get in its own way. A “whole of government” approach that includes export controls along with investment and supportive regulations should focus on this larger transformational goal where AI tools will play a central role.
The IFR restricts access to American-made AI technologies by both potential adversaries as well as countries that pose no real risk. A redesigned rule would expand the list of countries able to acquire AI hardware and services without licensing (but subject to transparency and catch-all requirements designed to reduce the risk of diversion). This would include countries like Switzerland, Poland, Brazil, India, Singapore, and Israel. These countries do not pose any military risk to the United States.
A redrafting of this IFR would create two country groups: one for high risk countries like China, Russia, and others in D:5 and another for everyone else. If a country that does not pose a risk of diversion or to national security wishes to buy GPUs and it is subject to transparency requirements, there is no reason for the US to block the sale. Diversion is a problem that the US has dealt with successfully the past by imposing restrictions on end-use and end-users and dedicating enforcement resources to ensure these are observed. The task of export controls is not to let the US manage another country’s acquisitions and it is unlikely that agencies in Washington are better judges of global demand for AI.
We do not want to sacrifice market leadership to prevent the potentially (and likely unavoidable) use of AI by a few hostile countries. Unavoidable, because there are adequate alternatives and substitutes to the products and services controlled by the IFR, making it possible for countries to circumvent it. The US cannot expect to have a monopoly on AI, but it can expect to lead the market and in innovation if its companies are permitted to do so.
Managing Diversion Risk
This proposed change in country groups would need to be buttressed by some kind of regular reporting to BIS on exports (on destination and quantity). It would also need to be accompanied by stringent measures to safeguard again diversion. The precedent is the export control designed for non-proliferation, which rely on the private sector. This rule would benefit from rebalancing economic potential and political importance against the potential risk of diversion, to give greater weight to the former.
A new or rewritten rule could be buttressed by “Know Your Customer” (KYC) requirements. KYC rules ensure that sensitive technologies do not fall into the wrong hands. KYC has been used for years to prevent exports of proliferation-related items to the wrong end-users or uses. KYC rules require that exporters screen customers against various government lists, including the Denied Persons List of individuals and companies barred from export privileges, the Entity List of parties posing a risk of diverting items to prohibited activities, the Specially Designated Nationals and Blocked Persons (SDN) List.
Extending this kind of “catch-all” control to cover AI, where a license is required only if an exporter “knows or has reason to believe” that an export will be used for prohibited purposes, or has been informed by the Department of Commerce (which can act as a conduit for the Intelligence Community) that an export would present an unacceptable risk would fully address national security concerns. The experience of KYC shows that a supporting industry of advisory and private intelligence firms will spring up to help companies identify risky transactions. Industry complaints about the difficulty of KYC are nothing new and given a choice between licensing and robust KYC, companies prefer the latter.
Diversion risk could also be mitigated by creating some kind of “trusted vendor” status for American companies that provide services and operating data centers overseas. Requirements for becoming a “trusted vendor,” reflect to a degree the previous Administration’s distrust of the market and preference for bureaucratic oversight and should not be a substitute for licensing. The better alternative for US market dominance is to assume that venders are trustworthy and create a framework of rules, transparency, and KYC requirements.
Thresholds
The BIS interim final rule on AI diffusion establishes a complicated set of thresholds designed to manage the export and transfer of advanced computing integrated circuits and AI model weights. These thresholds use technical specifications related to processing performance and training operations. They create geographical restrictions based on the tiered country structure and establish conditions for various license exceptions and validated end-user programs. These thresholds are unnecessary if the IFR is modified to use a “catch-all” approach, where licenses could be required for exports the US deemed to be risky. A simpler rule would ultimately be more effective and avoid damaging US tech leadership.
Use Regulation to Reinforce American Leadership, Not Undercut It
To ensure global AI market leadership, US policy and regulation should prioritize both innovation and market capture. The key metrics for a successful policy should be increased market share and revenue. Streamlining regulatory processes to favor rapid innovation to meet market demand and incentivizing private sector investment will accelerate global market penetration by US firms. The goal is to establish a dominant market presence, to drive global adoption of American AI technologies and take a leading position in the AI-driven economy.
America’s technological leadership is the product of technologies that are largely free from export control. Innovative American companies create products and services for a global market. They often face Chinese competitors who are the beneficiaries of China’s subsidized industrial policies and its predatory trade practices – the Huawei story exemplifies this. Europe has strong research and entrepreneurial capabilities and is now seeking to find ways to better turn these strengths into commercial success. America usually has better technology, but not always (China leads in EVs and batteries, for example), so leadership will not be conferred automatically.
The goal should be to design regulations to best capture global markets for tech and AI. This means allowing American companies to sell to the broadest possible market consistent with a realistic appraisal of risk, relying more on markets and the private sector for export decisions, limited only by government policy on the risk of transactions and diversion. This requires maximizing market opportunities to create the revenue streams that support and incentivize innovation, guided by the successful combination of public policy and private investment that makes America a tech powerhouse. Export regulations should be reoriented from seeking to prevent other countries from catching up and instead focus on ensuring that US companies remain in the lead
The commercialization of the internet is a good example of this approach. The technologies and services that make up the internet are largely not controlled for export. They form the backbone of the digital revolution that continues to reshape economies, and American companies lead in this. The decision by Europe to regulate the internet and insert government approvals for market-based decisions is one of the primary reasons it has fallen behind in digitalization. When it comes to AI, there are only a few countries (such as China) where the risks of an open approach outweigh the opportunities. American security is best ensured by export control policies that incentivize companies to lead globally, with decisions on investment and sales shaped by competition over market opportunities rather than micromanagement by government. The fundamentals for redrafting this rule are:
— To have only two country groups: Tier III and everyone else
— No performance thresholds
— Replace licensing with KYC and catch-all requirements and regular, post-export reporting for every country not in Tier III
— Expanded measure to restrict Chinese access to hardware or online AI services
— A recognition that the greater threat to American security comes from regulations that damage American innovation and market dominance, not the diffusion of AI technology
The crux of a rewrite of the IFR should be to rely more on markets and less on bureaucracy and focus on strengthening the US. China is a formidable competitor as we have learned from the Huawei and DeepSeek experiences, and export controls will not stop it from making advances in AI. Nor can the US expect to use an exaggerated and increasingly imaginary monopoly to govern a global AI ecosystem from Washington. The IFR actually harms the US in an AI and technology race with China and should be significantly amended or replaced to reflect this.