The growing interconnection of cyber-physical systems (CPS) across critical infrastructures has increased their exposure to complex and persistent cyber threats. Traditional defense mechanisms, which rely on centralized monitoring and static rule-based responses, are often insufficient to counter adaptive and coordinated attacks. This paper presents a novel framework termed Autonomous Network Guardians, a multi-agent artificial intelligence (AI) architecture designed to safeguard and self-heal CPS environments. The proposed system introduces cooperative AI agents that operate autonomously at the network edge, capable of identifying, isolating, and recovering from cyber disruptions without external intervention. Each agent employs an adaptive learning model to refine its defense strategy based on observed system behaviors, enabling proactive and context-aware responses. A consensus-driven decision layer ensures coordination among agents, preventing redundant actions and maintaining overall network stability. The self-healing mechanism integrates reinforcement learning and fault-tolerant control strategies to restore normal operations after anomalies are detected. Experimental evaluations demonstrate significant improvements in resilience, recovery time, and threat mitigation compared with conventional intrusion detection systems. Furthermore, the study emphasizes ethical considerations, ensuring transparency, reliability, and accountability in autonomous defense decisions. The proposed approach contributes a scalable and intelligent foundation for next-generation CPS security, aligning with the vision of fully autonomous and resilient network infrastructures. Future work will extend this framework toward large-scale deployment and cross-domain interoperability.

Keywords: Multi-Agent Artificial Intelligence, Self-Healing Networks, Cyber-Physical Systems Security, Autonomous Defense Frameworks, Intelligent Resilience Engineering, Decentralized Anomaly Recovery, Adaptive Cybersecurity Agents