In the race for artificial intelligence supremacy, the United States faces a critical challenge: maintaining our technological edge while preventing adversaries, particularly China, from exploiting our home-grown innovation. Fortunately, the bipartisan Enhancing National Frameworks for Overseas Restriction of Critical Exports (ENFORCE) Act addresses this challenge head-on by providing much-needed clarity to our export control framework.
The importance of AI to America’s national security and prosperity cannot be overstated. From enhancing our economic competitiveness to revolutionizing national defense, dominance in AI is poised to become the cornerstone of 21st-century power. Yet with great potential comes great risk. China is also investing heavily in AI capabilities, often with the explicit goal of undermining U.S. interests while exporting their model of digital authoritarianism to countries around the world.
Currently, the Bureau of Industry and Security (BIS) at the Department of Commerce has the authority to control the export of dual-use software and hardware technologies, including advanced semiconductors and manufacturing equipment. However, there's a glaring gap: BIS has unclear authority to restrict the export of AI systems themselves. This is because, unlike traditional software, the functionality of a neural network is fully captured by the “model weights” learned over the course of training—parameters that amount to a giant list of numbers.
This ambiguity is concerning given China's track record of intellectual property theft and its strategy of leveraging joint ventures with U.S. firms to acquire advanced technology. Recent reports have revealed underground networks smuggling Nvidia's AI GPUs into China, for example, circumventing U.S. export controls. This illicit trade not only undermines our national security but also gives China access to technologies that could accelerate its military capabilities.
The ENFORCE Act is designed to address these concerns without hampering U.S. innovation. It does this by clarifying the BIS’s authority to control the export of specific covered AI systems identified as essential to U.S. national security to foreign entities of concern. The Act itself does not impose any new controls or restrictions, but merely resolves that the BIS’s authority extends to the model weights of covered AI systems.
The Act defines "covered artificial intelligence system" as one that exhibits capabilities that pose a serious risk to national security, such as lowering barriers to develop weapons of mass destruction, enabling offensive cyber operations, or permitting evasion of human control or oversight. This is consistent with the BIS’s existing mandate, and helps to ensure that export controls on AI systems, if employed at all, remain narrowly targeted.
Critics might argue that any new export controls could hinder American companies' ability to compete globally. However, the ENFORCE Act takes a measured approach:
- It doesn't automatically impose new restrictions. Instead, it gives the president and BIS the tools to act when necessary.
- It focuses only on the most advanced, dual-use AI systems with clear national security implications.
- It allows for a nuanced approach, with the potential for controls to be tailored to specific threats.
Export controls of this kind are not particularly novel. Throughout the 20th century, the U.S. invested heavily in export administration to deny the Soviet Union and its proxies access to our most sensitive technologies, including military arms and munitions. Frontier AI systems are fast approaching similarly dual-use levels of capability, be it in the form of an offensive cyber weapon or a platform for designing deadly biological agents. The primary difference is that, unlike a ballistic missile, the most advanced AI systems will soon cost billions of dollars to produce while being essentially costless to move across borders.
The Chairman of the House Foreign Affairs Committee, Michael McCaul, put the challenge well: "As the CCP looks to expand their technological advancements to enhance their surveillance state and war machine, it is critical that we protect our sensitive technology from falling into their hands." To the extent this includes our most advanced and weaponizable AI systems, the ENFORCE Act is thus a major step in the right direction.