Artificial Intelligence and Global Security: Governing AI in Nuclear Weapons Systems

Abstract

Artificial intelligence (AI) has the potential to destabilize global security, and nuclear weapon systems represent a critical case study. Nuclear command, control, and communications (NC3) provides the technical infrastructure to authorize and enable the use of nuclear weapons. Virtually all nuclear-weapon states are either considering, pursuing, or actively integrating AI into NC3. This thesis examines how AI can introduce risks in NC3 and evaluates current global governance initiatives to mitigate risk. Examining AI in NC3 through the lens of risk-tiering, verification, and the feasibility of international governance offers key insights into the limitations of current AI governance frameworks. The inquiry seeks to evaluate the following question: What risks will arise from the incorporation of AI into NC3, and how can international communities seek to manage the risks of AI in nuclear weapons systems? The methodology incorporates findings from relevant literature, twenty expert interviews with individuals in academia, government, military, and the private sector, and two expert conferences. This thesis determines that the six most notable AI applications in NC3 span three risk tiers—lower risk: planning and research & development; moderate risk: monitoring, decision-making, and force direction; and the highest risk: launch authority. An acceleration paradox emerges: AI integration into NC3 can create an illusion of providing additional deliberation time, but in reality, AI speeds up decision-making and can be unreliable. As hypothesized, verification challenges limit the feasibility of multilateral agreements on AI in NC3, but the accelerated pace of decision-making and newfound unreliability of AI demands international action. While the “human in the loop” principle aims to prevent AI launch authority, it does not address the risk posed by low or moderate risk applications of AI that nuclear-weapon states will pursue, creating a governance gap for the most relevant applications of AI in NC3. Ultimately, this thesis argues that the most pressing governance gap for AI in NC3 exists in the moderate risk tier—AI-enabled monitoring, decision-making, and force direction—where nuclear-weapon states will implement AI applications. The most practical policy solutions to reduce this governance gap are to optimize public-private sector collaboration, generate rigorous testing systems, establish a shared code of conduct based on risk-tier frameworks, and lead multilateral dialogue to foster international norms on AI in NC3.