Explore how a tool-mediated LLM architecture provides formally verified stability for autonomous cyber defense, ensuring robust EDR policy optimization against adaptive AI adversaries.