Abstract

Critical infrastructure systems increasingly face sophisticated cyber-physical attacks capable of simultaneously compromising digital control networks and physical operational processes. Conventional rule-based and signature-driven security mechanisms are inadequate against advanced persistent threats, zero-day exploits, and coordinated multi-vector attacks. This paper proposes an autonomous artificial intelligence-based defense architecture for resilient protection of critical infrastructure sectors, including energy, water, transportation, and telecommunications. The proposed architecture integrates deep autoencoder based anomaly detection, LSTM-driven temporal behavior analysis, deep reinforcement learning for autonomous response selection, and multi-agent coordination with federated learning to enable real-time threat detection, adaptive response, and rapid system recovery. The system is evaluated using benchmark intrusion detection datasets, cyber-physical simulation environments, and real-world deployment case studies across water treatment facilities, electrical substations, and transportation networks. Experimental results demonstrate a detection accuracy of 97.3% with a false positive rate of 0.8%, while autonomous response mechanisms reduce mean time to detection by 73% and improve overall system resilience by 84% compared to traditional approaches. The architecture achieves sub-second detection latency and maintains high service availability under attack conditions. These findings indicate that autonomous AI-driven defense systems provide a scalable and effective foundation for securing modern cyber-physical infrastructure, offering significant improvements in resilience, responsiveness, and operational safety in increasingly connected critical environments.