A 1909 military test of an experimental hydroplane. American air supremacy, uncontested since 1945, defined the 20th century’s international security architecture. Source: Library of Congress.
Federal policymakers weigh historic tradeoffs in their current race to shape AI policy. Where they choose to focus will determine whether the U.S. achieves dominance in the decisive technology of the century — or cedes it to authoritarian adversaries.
The rapid evolution of artificial intelligence (AI) has placed it at the forefront of global technological competition, transforming it into a strategic imperative for national and economic security. It is crucial to position America as the global leader in AI development, deployment, and regulation to safeguard our citizens from adversaries who might exploit AI for malicious purposes. I warned in my 2021 book The Wires of War: Technology and the Global Struggle for Power that AI is the pivotal technology in U.S.-China competition and that the U.S. government and Silicon Valley must seriously partner to protect democracy from the autocrats looking to dismantle it.
Understanding the strategic imperative of getting AI policy right, Congress spent the last year debating the subject.1 The conversation in both Democratic and Republican circles is informed in large part by two basic truths:
1) Without a real partnership between the U.S. government and Silicon Valley, neither is fully equipped to protect democracy from the revisionist adversaries looking to pick it apart.
2) AI is the essential omni-purpose technology of the U.S.-China arms race, like steel in medieval warfare.
Both Democrats and Republicans see the need for a policy approach that values America’s global dominance in this new technology, with Senator Chuck Schumer recognizing “no technology offers more promise to our modern world than artificial intelligence” and House Speaker Mike Johnson asserting that “American companies are the world’s leaders in AI technology and we desperately want to maintain that status.”2
Many thinkers are also now seriously considering national security implications of AI policy, with competition with China a central focus. At this year’s Hill & Valley Forum, House Speaker Mike Johnson and other lawmakers cautioned against rushing to over-regulate AI.3 At the June 2024 Reindustrialize Conference in Detroit, House Majority Leader Steve Scalise and I repeated the need for a light-touch approach to AI regulation in a joint virtual address.
Over months of discussions with policymakers, 11 high-level principles have stood out as those most likely to define the next phase of the AI policy debate. Please note that these principles were not derived from discussions with any presidential campaign — rather they reflect the current state of AI policy in Congress.
There is remarkable similarity between the parties on the importance of getting the AI regulatory landscape right. The challenge now: where should lawmakers and policymakers focus limited time and attention? These 11 points of consensus and debate will yield the most consequential benefits for the country or produce the most adverse impacts.
1. A Sector-Based, End-Use Approach to AI Regulation
Developing an entirely new one-size-fits-all regulatory framework for AI would likely stifle innovation and fail to address the nuanced needs of different industries. Instead, the U.S. should adopt a sector-based, end-use-focused approach to AI regulation by applying and building upon the existing rules of each federal agency governing various sectors of our economy, from healthcare to finance to transportation.4 The risks and potential harms of AI vary greatly from sector to sector. The risks of self-driving cars, for example, are different from those of AI-powered financial trading or AI-assisted medical diagnoses. Existing regulatory principles and precedents can be an invaluable guide for addressing these, and every federal agency should begin exploring how existing laws on the books can be applied to AI. This keeps the focus on applying existing rules that protect public safety and privacy to AI without stifling innovation.
Areas of agreement: The most significant evolution in the AI regulation is the collapse in political support for a new, all-encompassing regulatory agency. Lawmakers are instead exploring ways to allow existing regulatory agencies to enforce existing laws and rules on a sector-by-sector basis as AI becomes more embedded in cars, hospitals, planes, and other areas of everyday life.5 This is the preferred approach of most in the startup and venture ecosystem and reduces the likelihood of burdensome regulations that stifle innovation.
Areas of debate: Democrats generally favor additional regulation, particularly around safety, bias, and workforce displacement, while Republicans remain more concerned about protecting free speech and cautious about stifling innovation.
2. Federal Preemption
Some states are considering state-level laws and rules focused on risk-mitigation in AI. California’s SB 1047 is the latest example. The law puts in place complex reporting requirements for developers who fine-tune models or develop models that cost more than $100 million to train. The law’s penalties are high, and its language is vague; the result will be uncertainty and paralysis, discouraging innovative entrepreneurship in this space. A federal preemption would help ensure uniformity and consistency across all states, drastically simplifying compliance for businesses, individuals, and enforcement agencies alike. Dealing with a single federal law is generally more efficient than navigating a patchwork of 50 state laws. This would reduce complexity, legal costs, and administrative burden for all stakeholders involved. Federal preemptions typically focus on issues that involve activities that cross state lines, such as commerce and telecommunications, and AI seems to fall in this category. Finally, some issues, like national security or certain aspects of healthcare, are often better addressed at a federal level to ensure a cohesive national strategy and response.
Areas of agreement: There is generally broad opposition in the technology world to the multiplication of state-level laws on artificial intelligence. There is also general agreement that AI is a highly strategic technology and should be regulated by institutions which can consider national security — the strict legal purview of the federal government.
Areas of debate: The jury is out on what the substance of a potential federal preemption or law could or should look like. If the debate to pass a federal preemption gains traction, some in Congress will likely attempt to broaden the bill to include topics around safety or other AI issues orthogonal to preempting state-level laws.
3. Removing Regulatory Barriers to Semiconductor Manufacturing
Advanced AI systems require significant compute power and secure access to semiconductors. While the United States has potential to produce these critical resources and power dominant U.S. AI systems, federal agencies like the Environmental Protection Agency (EPA) have issued regulations that impede the construction of new chip manufacturing facilities in the U.S., putting the country’s AI industry at a disadvantage.6 Eliminating these and providing a means to fast-track construction of chip fabs, whether through congressionally backed permitting reform or through the normal EPA permitting process, is a critical first step for the U.S. to quickly bolster its AI supply chain.
Areas of agreement: There is general bipartisan consensus around the need to accelerate the construction of chip fabs in the United States. Although the momentum for a bill to codify a fast-track process into law is lacking, a bipartisan majority supports a solution that fast-tracks chip fab construction through the normal EPA permitting process.7
4. Promoting Market Transparency and Benchmarking
AI models are currently a black box — their own builders are often not entirely aware of their full capabilities, let alone the general public.8 Mechanisms that allow investors and regulators to benchmark (and compare) the capabilities and potential risks of one model relative to another are important for building competition, trust, and accountability in this new trillion-dollar market. Just as Moody’s and Fitch rate the creditworthiness of borrowers, the United States should explore a similar type of benchmarking system for AI systems.
Areas of debate: There is currently no consensus on how to best approach benchmarking. Some have floated the idea of empowering a government agency or office to benchmark and audit AI models. Others have favored an industry-led approach where government-accredited private entities undertake benchmarking. Others have argued that the market already has private entities focused on benchmarking, such as HuggingFace.9
5. Exporting American Foundation Models as a Global Platform for AI Development
Like the Internet, AI is a platform technology on top of which thousands of new applications will be built. To ensure our long-term strategic advantage, America — and its AI companies — must offer a Free World alternative to China’s authoritarian AI platforms. In the past two years, Chinese authorities have released regulations to ensure that Chinese-built AI “adhere[s] to the correct political direction,” does not “disturb [the] economic and social order,” and “reflects core socialist values” — code for spreading and refusing to challenge CCP propaganda.10
Some of these regulations include sections targeting AI models developed and used outside Chinese jurisdiction, suggesting that the CCP may aim to export their AI regime.11 American-built AI is poised to have just as transformative an effect as the U.S.-built free and open global Internet and the U.S. dollar, which became the foundation on which much of the world has built its economy. Measures that restrict the global adoption of American AI technologies risk creating a global vacuum to be filled by Chinese AI companies. As with the Internet, most nations will face a choice in the years ahead between American-built AI embodying concepts of personal privacy, free speech, and intellectual property rights, and Chinese AI built for surveillance, censorship, and intellectual property theft.12
Areas of debate: There is an ongoing, hotly contested debate in the technology industry on finding the right balance between the competing priorities of making America the world’s AI platform and preventing U.S. adversaries from gaining access to America’s most advanced models. While some favor a complete and total “open source” approach to AI models, where everyone has the same unfettered access to a given AI model, others favor a Know-Your-Customer rule for the most advanced AI models, which could allow the government to surgically restrict access to those models for specific entities or countries of concern. Others have also floated the idea of selectively instating export control restrictions applied to certain models, on a case-by-case basis, if they exceed a certain threshold in size, compute or capability. This remains one of the most contested debates in the technology industry.13
6. Securing the AI Supply Chain
America’s AI supply chain must be capable of weathering the potential shocks of a deteriorating geopolitical climate. Therefore, the U.S. must incentivize the reshoring of supply chains for electronic hardware, including the production of semiconductors, and guarantee access to critical raw materials and machinery and other critical inputs. The U.S. must also ensure American AI companies train their largest and most advanced models on U.S.-based computing clusters rather than relying on other countries’ data centers. To accomplish these goals, it is critical for the federal government to continue efforts to selectively decouple our economy from China, particularly when it comes to semiconductor technology.14 Cognizant that decoupling in hardware manufacturing will take time, policymakers are quietly considering the idea of building strategic semiconductor stockpiles — as China has been doing for years.15
The U.S. must also retain a sufficiently robust private domestic computing capacity to ensure American AI companies can train their largest and most advanced models on U.S.-based computing clusters, should a geopolitical crisis cut off or compromise their access to computing servers in other overseas data centers. While there has been significant focus on expanding America’s chip building capacity through initiatives like the CHIPS and Science Act of 2022, relatively little attention has been brought to this risk of American AI companies becoming reliant on foreign large computing clusters to train the world’s most advanced models.16 As energy-rich nations like the UAE promote the co-location of large computing clusters near abundant sources of fossil fuel energy, the U.S. has an opportunity to harness its competitive advantages in energy to further expand its domestic computing capacity.
Areas of agreement: Members of both parties support the use of Section 301 tariffs on Chinese goods and support the use of additional levies to level the macroeconomic playing field with China.17
Areas of debate: Despite widespread support for Section 301 tariffs and other methods of supporting U.S. chip production, there is still debate over how high the tariffs should be and how broadly they should be applied.18 Moreover, little attention has been brought to harnessing AI models as new instruments of trade policy. This will likely be an area of greater debate in 2025.
7. Cheap, Abundant Energy
The global race to train increasingly complex models requires vast computational power, which in turn demands enormous amounts of cheap, abundant energy. Countries like the UAE are leveraging their access to vast energy reserves in order to reinvent themselves as global hubs in the AI supply chain with the construction of massive and computing clusters co-located near abundant sources of oil.19 The U.S. government should facilitate the exploitation of all forms of abundant low-cost energy, including oil, gas, and nuclear energy, to accelerate the private sector’s development of artificial intelligence and compete with countries like the UAE. The U.S. government should also encourage and accelerate the private sector’s deployment of emerging breakthroughs in nuclear energy, such as small modular reactors and fusion energy.
Areas of debate: Energy policy remains a mostly partisan issue. There remains significant debate over fast-tracking the permitting needed for new sources of oil, gas, and nuclear energy. Support for this program falls largely along party lines, with Republicans in favor and Democrats opposed.20
8. Protecting Americans Against Foreign Adversaries
Preventing adversaries from exploiting cutting-edge AI technologies is critical for national security. Serious controls on imports and exports of strategic technologies, robust cybersecurity measures, and strategic alliances are necessary to safeguard against malicious use of AI by our adversaries. Collaboration with allies to create a unified front against intellectual property theft and cyber threats will further secure America’s technological edge.
Areas of agreement: The Biden administration built on many of the Trump administration’s export controls and other measures to ensure America’s adversaries weren’t benefitting from American technology. As more lawmakers grow concerned about the near-term threat posed by Chinese technology companies, there will be a continued bipartisan consensus to expand and strengthen export controls as well as regulate capital flows into American companies. A growing concert of import control rules are taking shape to safeguard against risks of strategic technologies controlled by foreign adversaries, particularly the Chinese Communist Party. The TikTok divestiture passed Congress with an overwhelming majority of both parties.21 Scrutiny over other Chinese-controlled products like Huawei, Hesai lidars, DJI drones are likely to persist and deepen.
Areas of debate: Most of the scrutiny to date has focused on flows of capital and hardware such as advanced semiconductors and ASML machinery to our adversaries; little scrutiny has focused on imports of AI-powered tools from our adversaries. Recent breakthroughs in AI-powered robots, particularly intelligent humanoid robots, will become a flashpoint of public scrutiny and debate in 2025. Little scrutiny is also being placed on the surge in venture capital investments from China in cutting-edge American AI companies. These investments often come in conjunction with the allocation of observer seats to the company’s board of directors and information rights, making the protection of sensitive intellectual property extremely challenging.22
9. Recruiting the World’s Best Talent
As the Manhattan Project made evident in the 1940s, attracting the best and brightest minds is critical for developing emerging, strategic technology like AI.23 Unfortunately, we retain an outdated immigration system that dramatically curbs our ability to bring in the highest-skilled immigrants who can address the shortage of workers in the AI industry and contribute to American AI dominance. Modest reforms to our immigration system, such as allowing the government to prioritize visas for high-skilled talent without changing the overall caps on the number of legal immigrants allowed into the country, would dramatically benefit our technology workforce and companies that rely on high-skilled labor.24 Others have suggested a new version of the Space Race’s “Operation Paperclip” aimed at identifying, targeting and recruiting the world’s most talented minds on AI development and semiconductor manufacturing and quietly negotiating their immigration to the United States.25
Areas of agreement: There is a striking degree of bipartisan agreement around the need to make it easier for the world’s best research scientists in strategic fields like AI to work in the United States. Unfortunately, Congressional action on this issue has been stymied by broader disagreements on comprehensive immigration reform and the current crisis at the Southern border. Refusals to decouple reforms affecting high-skilled legal immigrants from other immigration issues have stalled progress.26 Nevertheless, new opportunities to make reforms to the way in which the United States attracts high-skilled immigrants may arise in the next legislative term, fueled by the growing bipartisan sense of urgency around the need to compete with China.
Areas of debate: The debate over immigration remains extremely polarized. Congress has made little effort to address immigration issues in a piecemeal approach and instead sought to address immigration issues through larger, comprehensive bills. This approach, coupled with disagreements over how to handle illegal immigration, Dreamers, and the Southern border, have reduced the likelihood of Congressional action.
10. Modernizing Civilian Government Services
AI has the potential to revolutionize the delivery and operations of civilian government services. By integrating AI, the government can enhance efficiency by an order of magnitude, eliminate waste, fraud, and abuse, and significantly improve the delivery of public services. From improving how the FAA manages our crowded airspace, to better allocating health and human services across different communities, to dramatically compressing administrative process delays, there are endless opportunities for AI to bring about a more responsive, leaner, and cost-effective government.27
Areas of agreement: There is bipartisan recognition that AI can transform the way the government delivers services to the American people. However, bureaucracy is always slow to move — and even slower to adopt and incorporate new technology. Any effort to quickly integrate AI across the federal government will require committed, top-down Presidential leadership that drives change across agencies.
Areas of debate: The adoption of breakthrough AI technologies by federal government agencies will face skepticism from a minority of voices on both sides of the aisle. On the right, deficit hawks will be weary of any meaningful spending measure without ironclad assurances of spending-reductions over time or in other areas. On the left, some will have reservations around new programs that commercially benefit private enterprises.
11. Modernizing the Military and Defense Industrial Base
Modernizing the military through AI is critical for maintaining America’s strategic edge over our adversaries. AI can enhance decision-making, operational efficiency, and logistical support within the military. For example, AI could enable predictive maintenance and advanced simulation for training and strategy development.28 The military must focus both on modernizing its software systems and on procuring the private sector’s best dual-use AI-powered technologies to deploy on the battlefield. Ongoing conflicts in Ukraine and the Middle East demonstrate how AI is enabling powerful and rapid synchronization of all logistical, military and intelligence capabilities. Breakthroughs in autonomous robotic systems, ranging from drones to humanoid robots, are also slowly revealing the new shape of war in the decades to come. AI and an explosion in the production of dexterous robots in the United States and China are poised to create a once-in-a-century inflection point in military and industrial history. Intelligence and weapons-enabled robots in land, air, sea and space will likely be controlled and synchronized by AI, allowing them to learn, refine and improve warfighting capabilities on the fly and in an exponential way. The development of military AI is an arms race, chiefly between the U.S. and China, which has already begun. While the exact end-state of general military AI is unknown, there is little debate that military AI is a step-change, and the outcome of the global AI race could result in a shift in the global balance of power.
Additionally, as ongoing conflicts in Europe and the Middle East have exposed vulnerabilities across our defense industrial base, the military should capitalize on AI’s ability to revolutionize manufacturing by introducing intelligent manufacturing processes. AI could accelerate scaling, exponentially improve process know-how, and compress the time delays of each stage of the production process to combat U.S. military manufacturing shortfalls. Intelligent robots capable of human-like dexterous manipulation are also poised to increasingly relieve labor-shortage constraints in the United States related to highly difficult, repetitive and technical tasks.29
Throughout history, major technological step-changes have shifted the global balance of power. The United States is locked in a technological arms race with China, and AI will be decisive in this competition. To secure the nation’s future in an increasingly competitive world, we must enact an agenda for American AI dominance. These points of bipartisan agreement and debate will be foundational to achieving that dominance, protecting U.S. national security, and driving economic prosperity.
References
1 Emily Gaines Buchler, “AI Regulation Necessary to Address Potential Risks, Key Senators Say,” The Hub, February 29, 2024.
2 Following Historic AI Insight Forums over the Past Year, Leader Schumer, Senators Rounds, Heinrich, & Young Reveal Bipartisan Roadmap for Artificial Intelligence Policy in the United States Senate (Washington, D.C.: Office of U.S. Senator Chuck Schumer, 2024); “Hill & Valley Forum: Mike Johnson,” YouTube, May 9, 2024.
3 “What We’re Hearing: Speaker Johnson on AI,” Axios, May 2, 2024.
4 Richard Manfredi, Artificial Intelligence Review and Outlook – 2024 (Los Angeles: Gibson Dunn, 2024).
5 Hadrien Pouget and Matt O’Shaughnessy, Reconciling the U.S. Approach to AI (Washington, D.C.: Carnegie Endowment for International Peace, 2023).
6 Justine Calma, “How the next Generation of Semiconductor Factories Kicked up a Fight over Environmental Review,” The Verge, October 13, 2023.
7 Ibid.
8 Steven Levy, “AI Is a Black Box. Anthropic Figured out a Way to Look Inside,” Wired, May 21, 2024.
9 Ryan Heath, “Makers of Leading AI Models Can’t Agree on ‘Responsibility’ Benchmarks,” Axios, April 16, 2024.
10 Matt Sheehan, China’s AI Regulations and How They Get Made (Washington, D.C.: Carnegie Endowment for International Peace, 2023).
11 Anna Gamvros et al., “China Finalises Its Generative AI Regulation,” Data Protection Report, July 25, 2023.
12 Kaan Sahin, The West, China, and AI Surveillance (Washington, D.C.: Atlantic Council, 2020).
13 David Evan Harris, “Open-Source AI Is Uniquely Dangerous,” IEEE Spectrum, January 12, 2024.
14 Jacob Helberg and Enes Kanter Freedom, “The United States Can’t Afford to Stay Entangled with China,” Foreign Policy, June 1, 2022.
15 “China Stockpiles Chips, Chip-Making Machines to Resist U.S.,” Bloomberg, February 2, 2021.
15 Brady Helwig and P.J. Maykish, “We Need a Moonshot for Computing,” MIT Technology Review, December 28, 2023.
17 Joseph Majkut et al., Experts React: Energy and Trade Implications of Tariffs on Chinese Imports (Washington, D.C.: CSIS, 2024).
18 Ibid.
19 Federico Maccioni, “Exclusive: UAE and US to Seal More AI Deals, UAE Minister Says,” Reuters, May 21, 2024.
20 James Broughel, “On Permitting Reform, Time Works in Republicans’ Favor,” Forbes, August 1, 2023.
21 David Shepardson, “US Senate Passes Tiktok Divestment-or-Ban Bill, Biden Set to Make It Law,” Reuters, April 24, 2024; David Shepardson, “US House Passes Bill to Force Bytedance to Divest Tiktok or Face Ban,” Reuters, March 14, 2024.
22 “Sequoia Capital’s Roelof Botha and Jacob Helberg on AI and National Security,” YouTube, June 25, 2024.
23 David N. Schwartz, “The Immigrants Who Saved America — and the Rest of the Free World,” Washington Post, December 2, 2017.
24 Chris Parsons Dany Bahar, et al., Smarter Immigration Policies Could Help Alleviate the Semiconductor Shortage (Washington, D.C.: Brookings Institute, 2022).
25 Shyam Sankar, “Time for a preemptive Operation Paperclip II,” X (formerly Twitter), June 19, 2024.
26 William A. Galston, et al., The Collapse of Bipartisan Immigration Reform: A Guide for the Perplexed (Washington, D.C.: Brookings Institute, 2024).
27 Niklas Berglind, Ankit Fadia, and Tom Isherwood, The Potential Value of AI-and How Governments Could Look to Capture It (New York: McKinsey & Company, 2022).
28 John Burnham, “The Use of AI in War Games Could Change Military Strategy,” Military.com, April 22, 2024.
29 DOD Is Taking Steps to Shore up Industrial Workforce (Washington, D.C.: U.S. Department of Defense, 2023).