Explainable Autonomous Cyber Defense using Adversarial Multi-Agent Reinforcement Learning

ArXi:2604.04442v1 Announce Type: cross Autonomous agents are increasingly deployed in both offensive and defensive cyber operations, creating high-speed, closed-loop interactions in critical infrastructure environments. Advanced Persistent Threat (APT) actors exploit "Living off the Land" techniques and targeted telemetry perturbations to induce ambiguity in monitoring systems, causing automated defenses to overreact or misclassify benign behavior as malicious activity.