AI-Based Cross-Layer Vulnerability Management for Cloud-Native Systems

Cloud-native systems introduce complex and dynamic attack surfaces driven by microservice architectures, identity-centric access models, and continuously evolving runtime behavior. Traditional vulnerability management approaches rely on static severity scoring and siloed security signals, resulting in excessive alert noise and limited insight into real-world exploitability.This paper presents a novel AI-driven cross-layer vulnerability management framework that models exploitation risk as an evolving behavioral process rather than a static property of individual vulnerabilities. The proposed approach represents cloud-native environments as temporal cross-layer graphs and applies self-supervised sequence learning, graph-based risk propagation, and anomaly detection to identify multi-stage exploitation paths spanning application, identity, runtime, and endpoint layers.Unlike conventional prioritization techniques, the framework learns how risk propagates across layers over time and adapts to previously unseen attack strategies without reliance on labeled exploit data. A simulated Kubernetes-based case study demonstrates substantial alert noise reduction while preserving exploitation-relevant signals, outperforming severity-based and rule-based correlation approaches.