As threats evolve across cloud and edge surfaces, modern cybersecurity demands adaptive, policy-aware AI systems that can detect, reason, and respond autonomously. In this session, we present a security-centric, federated architecture built on Kubeflow. Lightweight ONNX models run on K3s-based edge clusters to analyze Falco logs and detect anomalies locally, minimizing exposure. Only filtered signals and key features are shared to uphold compliance with GDPR, HIPAA, and zero trust principles. In the cloud, KubeFlow orchestrates a GraphRAG pipeline powered by vLLM and Knowledge Graph reasoning aligned to NIST and MITRE ATT&CK. This architecture supports detection, intelligent recommendations and rule automation—such as policy and rule updates—deployable back to the edge. Designed for high-stakes environments like healthcare, OT, and critical infrastructure, this session equips security and DevSecOps teams with practical tools to implement self-improving, compliance-aligned AI defenses.