America Needs a Comprehensive National Security Strategy for Artificial Intelligence

AI Dominance Means National Security

Recent developments in artificial intelligence (AI) leave little doubt: this is no ordinary consumer product. AI systems are now capable of enabling sophisticated cyber-attacks by uncovering thousands of software vulnerabilities in infrastructure thought to be secure. The latest AI systems are so capable of cyber offense that AI companies no longer give the public access to their best models.

These product releases are now national security events.

But it would be a mistake to solely fixate on AI’s cyber capabilities. AI will transform the national security enterprise and will become increasingly critical for both scientific and weapons research and development (R&D). Drones and robotics will transform the battlefield. AI will accelerate biotechnology, including the synthesis of novel biological weapons no responsible state would pursue. On the current trajectory, we are not prepared to prevent these capabilities from proliferating to our adversaries, nor are we prepared to address the stark reality that these capabilities may threaten the homeland.

The Trump administration has taken bold actions to bolster American AI leadership, calling for AI dominance to be the north star of AI policy. President Trump defined it broadly:

“It is the policy of the United States to sustain and enhance America's global AI dominance in order to promote human flourishing, economic competitiveness, and national security.”

However, recent AI capabilities demand a re-examination of the national security component of AI dominance. To truly dominate, America needs a national security strategy for AI. The US government (USG) must reject the “safetyism” and “doomerism” of previous approaches that would have overregulated AI, recognize the ways in which AI will determine the geopolitical future, and act to meet the challenge.

AI Will Fundamentally Transform National Security

In the process of crafting a National Security Strategy for AI, we must consider the fundamental factors driving AI’s importance to the national security enterprise:

First, AI is rapidly developing national security-relevant capabilities on uncertain timelines. Today’s AI systems are critical for military decision support and can autonomously discover and exploit cyber vulnerabilities. Tomorrow’s systems will likely be far more capable, yet it is difficult to predict exactly when critical capabilities will be developed. AI will likely become capable of significantly accelerating weapons R&D with little to no human involvement, but we do not know whether this capability will be reached in one year or ten years. Similarly, AI will likely streamline the creation of novel bioweapons, but we do not know when the first AI-enabled pandemic will spread. This uncertainty extends beyond timelines. As increasingly capable AI agents are deployed into higher-stakes settings, the potential for large-scale malfunction grows in ways that are themselves difficult to predict.

The USG must prepare for these contingencies.

Second, there is no guarantee that America will continue to lead in AI innovation and adoption. Given AI’s potential to rapidly shift the national security landscape, America must possess the best AI capabilities and the ability to deploy them across military and civilian sectors. However, dense regulations (such as energy-related permitting) and internal red tape (such as burdensome procurement policies) may squander our current lead.

Third, we cannot assume that defenders will keep pace with attackers. In certain domains, advanced AI capabilities may offer attackers a fundamental advantage over defenders. AI will likely accelerate the development of at least some such “offense-dominant” technologies, such as biological weapons in violation of the Biological Weapons Convention. It will likely be far costlier to defend against a lab-designed deadly virus than it will be to manufacture one.

Fourth, advanced AI capabilities tend to proliferate widely, including to hostile states, terrorists, and criminals. It takes an average of three months for open-weight AI systems to catch up to the capabilities of frontier AI systems. The proliferation and use of open-weight AI systems cannot be limited by either AI developers or the USG. This means that once powerful capabilities are discovered, defenders have only a brief window before a wide variety of malicious actors can easily obtain these capabilities. For some capabilities, threat actors may need little compute to threaten the American homeland.

Finally, American AI is not secure from theft and sabotage by adversaries. As AI becomes more critical for national security, foreign adversaries will be increasingly incentivized to steal and sabotage American AI technology. We are not prepared for such threats. No major AI company’s security posture is sufficient to prevent the theft of its frontier AI model technology by sophisticated nation-state adversaries. Once AI technology is stolen, it is trivial for the attacker to fine-tune away any safeguards and immediately weaponize the AI’s capabilities against the American people. Adversaries are also likely capable of sabotaging our most valuable AI systems by clandestinely inserting malicious, undetected behaviors into our most advanced AI.

Toward a National Security Strategy for AI

To address AI’s transformational impacts on national security, America needs a national security strategy for AI. Such a strategy should address: (I) promoting situational awareness of AI in the USG; (II) ensuring American AI leadership; (III) denying our adversaries access to AI; and (IV) preventing AI-enabled emerging threats.

Pillar I: Situational awareness of AI in the national security enterprise

To better inform all future decisions about a technology as complex and high-stakes as AI, the USG needs deep expertise in the technology and an accurate sense of its trajectory. This must include:

• Achieving real-time awareness of AI's national security-relevant capabilities so that the USG can quickly take necessary actions before emerging threats can no longer be addressed;

• Bringing sufficient AI talent and expertise into the federal government to understand complex information about AI; and

• Assigning agency leads to conduct independent testing and evaluation of frontier AI.

Pillar II: Ensuring America leads in AI innovation and adoption

This administration has taken unprecedented and necessary steps to accelerate American AI. Yet, given AI’s expanding importance to national security, further action is needed. A national security strategy for AI should have:

• A whole-of-government approach to accelerating AI adoption, including software and hardware;

• A robust domestic supply chain for physical actuators, like robotics and drones, lest we become dependent on our adversaries for the weapons of war; and

• Actions to maintain and widen our lead through AI infrastructure deregulation.

Pillar III: Denying our adversaries access to advanced AI

As advanced AIs become cyber weapons and engines of military R&D, we cannot afford to supply them or their ingredients to those who seek to harm Americans—whether sold willingly or stolen because of insufficient security. Denying access will require a strategy that:

• Secures the entire AI supply chain from theft, industrial espionage, and complex cyber operations;

• Strengthens export control rules and enforcement on physical chips, cloud computing resources, and semiconductor manufacturing equipment; and

• Develops a toolkit to track, assess, and potentially disable foreign projects that develop dangerous capabilities.

Pillar IV: Preparing for and preventing AI-enabled emerging threats

We must plan to become resilient to AI-enabled national security threats—both known and unknown. Doing so must include:

• Developing robust plans of action for responding to AI-enabled cyber crises;

• Analyzing and developing plans of action for AI-enabled bioweapons crises; and

• Requiring foreign adversaries to prove, using technical verification methods, that they are not using AI to threaten American lives.

AI is the newest frontier in national security. There is a narrow window for the federal government to shape AI for American dominance. Before us, the Cold War generation contained unfathomable risk through strategy and bold action. Our task is no smaller, though our window is smaller. Now is the time to act.