S. Paul
National Laboratory of the Rockies, Colorado, United States
Keywords: Energy resilience, AI assisted cybersecurity, Critical infrastructure protection, Compliance validation, Safety governed autonomy
Modern military mission continuity relies on Operational Technology (OT) and Industrial Control Systems (ICS) powering defense installations and critical infrastructure. As connectivity increases, security teams require repeatable, non-disruptive methods to discover vulnerabilities and validate defenses. Traditional AI-assisted security tools operate with unrestricted autonomy, posing severe reliability risks to safety-critical environments. To address this, we present a governed, multi-agent AI framework designed specifically for secure vulnerability discovery and compliance planning in power grids and defense installations. Unlike hazardous autonomous testing tools, this framework embeds strict authorization constraints, human approval gates, operational safety policies, and mission-aware stop conditions directly into its reasoning engine. Operating via six specialized agents (including Safety & Compliance, Evidence Grounding, and Planning), the system evaluates technical telemetry against cyber-threat intelligence to identify which weaknesses carry actual operational consequences. It plans controlled validation actions aligned with NERC CIP expectations and translates evidence into traceable, prioritized resilience recommendations. Cyber range testing confirms the framework can safely process raw observations and generate auditable defense insights without exposing critical mission networks to operational downtime.