Opportunities for the Application of Artificial Intelligence in Physical Security Risk Assessment

The increasing complexity and evolving nature of security threats to nuclear facilities, particularly nuclear reactors, necessitate a robust and efficient approach to physical security risk assessment.  This paper explores the integration of Artificial Intelligence (AI) into the existing frameworks of nuclear security risk assessment, focusing on kinetic attacks on facilities such as advanced reactors, including Small Modular Reactors (SMRs) and Microreactors (MRs).  The risk in nuclear security is defined as a function of threat, vulnerability, and consequences, with risk assessment being a systematic process of identification, estimation, analysis, and evaluation to inform decision-making. The application of AI offers several advantages in enhancing nuclear security risk assessments.  AI can enable more comprehensive and rapid analyses across a wide range of risk variables, uncover previously undetected vulnerabilities, and keep pace with evolving security technologies and situations, such as during armed conflicts.  Notably, AI's potential to assist in large-scale advanced reactor deployments and rapidly changing security scenarios is particularly valuable.  Additionally, AI support to existing, proven analysis tools can help maintain the regulatory authorities’ capacity to review and approve security plans, while addressing concerns related to AI’s "black box" nature. The discussion highlights the significance of Force-on-Force (FoF) modeling and simulation tools in nuclear security risk assessment, with a specific focus on tools developed by Sandia National Laboratories, Idaho National Laboratory, and ARES Security Corporation.  These tools, including AVERT Physical Security and EMRALD, are crucial for simulating physical security scenarios and evaluating the performance of Physical Protection Systems (PPS).  The integration of AI with these tools can support analysts, traditionally performing these tasks manually, enhancing efficiency and effectiveness. The paper argues for a stepwise integration of AI, starting with its application to support established FoF Modeling & Simulation tools.  This approach ensures that the core analysis remains grounded in proven methodologies, facilitating regulatory review and mitigating the opacity associated with AI processes.  The security and safety analysis work using AVERT-PS and EMRALD, and subsequent developments, provides a foundation for this AI application. Furthermore, the paper examines the unique challenges and opportunities presented by advanced reactors, especially MRs, due to their envisioned deployment lifecycle and diverse deployment scenarios.  While the operational phase might dominate the lifecycle, other phases pose novel security challenges.  The potential application of AI in these contexts opens new avenues for enhancing physical security risk assessments, ensuring the safe and secure operation of nuclear facilities in an increasingly dynamic and technologically sophisticated security landscape.