Nuclear Security in the Age of Artificial Intelligence

Abstract

The concept of artificial intelligence is fast changing the principles of nuclear security in the world. Nuclear security no longer focuses on the protection of the physical and control of the material, but includes cyber resilience, algorithmic decision-making, and more autonomous systems. AI is already used to contribute to such important areas as predictive maintenance of reactors, anomaly detection in the fuel-cycle facilities, and data fusion to nuclear command, control, and communication networks. Such capabilities will be able to reinforce the early warning, verification, and non-proliferation efforts. Nevertheless, they are also becoming more vulnerable as they are being adopted. Algorithms' effectiveness, data manipulation, adversarial interaction, and automation bias endanger to reduce human control and enhance the chances of error or unintentional growth. With AI spreading among the state and non-state actors, the dynamics of deterrence are increasingly rapid, unpredictable, and unstable.

This paper argues that AI has the potential to serve as a force multiplier in nuclear security, its infiltration of sensitive nuclear systems as it currently stands with a weak ethical, legal, and technical frameworks can be counterproductive to strategic stability. The paper analyzes the impact of AI on nuclear governance, deterrence, and operational planning by means of comparative analysis of great nuclear powers and other significant actors in the region. It outlines the areas of critical governance deficiencies in what is known as the regimes, specifically the Nuclear Non-Proliferation Treaty, the Convention on the Physical Protection of Nuclear Material, and the United Nations framework in general. The conclusion made in the paper is that an efficient AI-nuclear governance has to focus on transparency, human-in-the-loop control, and algorithmic accountability to make sure that AI enhances, and does not disrupt the global nuclear order.