CVE-2026-78427 Overview
CVE-2026-78427 is a policy bypass vulnerability in the NeuVector admission webhook. The webhook silently excludes containers from policy evaluation when their image path matches one of three hardcoded service mesh sidecar images. Because the image path is fully controlled by the workload author, any user able to deploy workloads can evade admission deny rules by renaming their image path to match a whitelisted sidecar. The flaw is categorized as reliance on untrusted inputs in a security decision [CWE-807]. It affects Kubernetes environments where NeuVector enforces admission control policies.
Critical Impact
Authenticated Kubernetes users with workload deployment rights can bypass NeuVector admission deny rules and deploy otherwise-blocked container images.
Affected Products
- NeuVector container security platform
- SUSE-distributed NeuVector packages
- Kubernetes clusters using NeuVector admission control
Discovery Timeline
- 2026-09-17 - CVE-2026-78427 published to NVD
- 2026-09-17 - Last updated in NVD database
Technical Details for CVE-2026-78427
Vulnerability Analysis
NeuVector integrates with Kubernetes as a validating admission webhook. When a pod specification arrives, the webhook evaluates each container against configured admission policies, including deny rules that block non-compliant workloads. To avoid interfering with service mesh traffic interception, NeuVector maintains a hardcoded list of three sidecar image paths that are exempted from policy evaluation.
The check compares the image path field of the container against this static allowlist. If the strings match, the container skips policy evaluation entirely. The webhook does not verify image digests, registry provenance, or signatures before granting the exemption.
An attacker who can submit pod specifications to the cluster can set the image field of a malicious container to one of the three whitelisted paths. The container then bypasses NeuVector admission deny rules while still pulling and running attacker-controlled image content from a registry the attacker controls.
Root Cause
The root cause is a trust boundary violation. NeuVector treats the workload-supplied image path as an authoritative identifier for a trusted sidecar. The image path is an untrusted input under attacker control, yet it drives a security decision that disables policy enforcement. This maps directly to [CWE-807], reliance on untrusted inputs in a security decision.
Attack Vector
An authenticated user with permission to create pods, deployments, or other workload objects crafts a manifest whose container image path matches one of the three hardcoded sidecar images. The manifest can reference an attacker-controlled registry, so the actual image content is arbitrary. On submission, the NeuVector webhook matches the path, skips admission evaluation, and the pod is admitted regardless of active deny rules.
Exploitation requires cluster access with workload deployment rights and does not require elevated privileges beyond that. No user interaction is required. See the GitHub Security Advisory GHSA-78r4-3wfq-r2xm for vendor detail.
Detection Methods for CVE-2026-78427
Indicators of Compromise
- Pod specifications referencing the hardcoded NeuVector sidecar image paths from registries other than the expected service mesh control plane.
- Admission audit log entries showing containers admitted without a corresponding NeuVector policy evaluation record.
- Running containers whose image path claims to be a service mesh sidecar but whose command, entrypoint, or process tree does not match the expected sidecar behavior.
Detection Strategies
- Compare the image path in admitted pods against the actual image digest and originating registry, alerting when a sidecar-named image is pulled from an unexpected source.
- Correlate Kubernetes audit logs with NeuVector webhook decision logs to identify workloads admitted with no policy evaluation trace.
- Baseline sidecar container behavior and flag deviations such as unexpected network destinations, file writes, or child processes.
Monitoring Recommendations
- Enable full Kubernetes API server audit logging for pods, deployments, and related workload resources with image path capture.
- Ingest admission controller and container runtime telemetry into a central data lake for correlation across namespaces and clusters.
- Alert on new image registries appearing in workload manifests that reference known sidecar names.
How to Mitigate CVE-2026-78427
Immediate Actions Required
- Upgrade NeuVector to the fixed version identified in GHSA-78r4-3wfq-r2xm.
- Restrict create and update permissions on workload objects to trusted service accounts and users.
- Audit existing workloads for containers using the hardcoded sidecar image paths from unexpected registries.
Patch Information
Refer to the GitHub Security Advisory GHSA-78r4-3wfq-r2xm and the SUSE Bug Report for CVE-2026-78427 for fixed versions and vendor guidance. Apply the vendor-supplied update that removes reliance on image path strings for sidecar identification.
Workarounds
- Enforce image provenance with a second admission controller such as a policy engine that validates registry and signature independent of NeuVector.
- Constrain allowed image registries at the cluster level so workloads cannot reference arbitrary sources for sidecar-named images.
- Apply Kubernetes RBAC to limit who can create pods with arbitrary image specifications, particularly in production namespaces.
# Configuration example
# Restrict allowed image registries with a Kyverno or Gatekeeper policy
# so that sidecar-named images can only be pulled from the trusted mesh registry.
# Refer to your policy engine documentation for the exact syntax.
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
