Autonomous Red Team AI: LLM-Guided Adversarial Security Testing

Abstract

This technical report describes an architecture for autonomous penetration testing using LLM-guided agents operating within Kubernetes-isolated environments. The system combines RAG knowledge bases with OODA-loop decision cycles, enabling systematic vulnerability discovery while maintaining strict NetworkPolicy isolation.