The vulnerability arises when Cilium’s L7 proxy allows traffic between pods and local L7 services on the same node to bypass Kubernetes NetworkPolicy enforcement under certain configuration conditions. With per‑endpoint routing enabled and BPF Host Routing disabled, traffic that should be filtered by Ingress NetworkPolicies is instead routed directly to the service backend. Attackers can exploit this by creating or compromising pods to send traffic that bypasses policy restrictions, potentially exposing workloads to unauthorized data flows or disrupting communication integrity. The weakness corresponds to improper access control (CWE‑284).

Affected deployments involve the open‑source Cilium networking solution prior to version 1.17.14, 1.18.8, or 1.19.2. The issue manifests when cloud IPAM configurations such as the Cilium ENI mode on Amazon EKS, AlibabaCloud ENI, Azure IPAM (excluding AKS BYOCNI), or various GKE deployments enable per‑endpoint routing. The most common affected environment is Amazon EKS when Cilium ENI mode is used, but any Kubernetes cluster that automatically enables per‑endpoint routing while disabling BPF Host Routing can be vulnerable.

The CVSS score is 5.4, indicating a moderate risk, and there is no EPSS data available; the vulnerability is not currently listed in CISA’s KEV catalog. Exploitation requires that per‑endpoint routing be active and BPF Host Routing be turned off—conditions that are often true for cloud‑provider deployments that auto‑enable per‑endpoint routing. Because no official workaround has been verified, the primary mitigation is to upgrade to a patched release or, if upgrading is not immediately possible, to disable per‑endpoint routing, acknowledging that this may interfere with existing connections on some cloud platforms.