Zero Trust Enforcement Through AI-Driven Isolation in Multi-Tenant Kubernetes Clusters on VxRail and AWS-EKS

Modern hybrid Kubernetes environments face unprecedented security challenges from lateral threat movement in multi-tenant architectures. This paper presents ZephyrGuard – an AI-driven Zero Trust enforcement system that dynamically isolates workloads across VMware VxRail on-premises and AWS EKS cloud clusters. Our framework integrates: 1) A federated LSTM autoencoder for real-time trust scoring using behavioral (network flows, process trees) and contextual (tenant sensitivity, CVE exposure) metrics, 2) A policy synthesis engine generating micro-segmentation rules via OPA/Rego with eBPF runtime enforcement, and 3) Cross-platform integration with VMware Tanzu and AWS GuardDuty. Testing on a 120-node hybrid cluster showed 97.3% F1-score in threat detection, 18ms P99 isolation latency, and 43% reduction in attack surfaces versus static policies. Comparative analysis demonstrated 5.9× faster incident containment than Tetration and 68% lower false positives than Illumio Core, establishing a new paradigm for adaptive security in cloud-native ecosystems.