Human-Aware Multi-Agent Cyber Defense for Distributed Nuclear Operational Technology Systems

Abstract

Operational Technology (OT) systems used in small modular reactor deployments (SMR) are becoming increasingly interconnected, increasing exposure to cyber threats that can affect tightly coupled physical processes. These systems operate under strict real-time and reliability requirements, which limit how security mechanisms can be deployed and evaluated. We present a multi-agent system for cybersecurity analysis across distributed nuclear OT environments in SMR fleets. The system is built on a simulation-based framework and uses a supervisory agent to coordinate specialized agents for vulnerability analysis and remediation. Agent interactions are restricted through controlled interfaces, and safeguards are applied to prevent unsafe actions during analysis. The system is designed to support continuous inspection of simulated environments while maintaining separation from operational systems. Human operators remain involved in decision-making when actions affect system behavior, enabling security analysis to be performed without introducing risk to live infrastructure.