On this page (7 sections)
A critical flaw in Kyverno, the Kubernetes policy engine, lets a tenant break out of the namespace their policy is supposed to be confined to. According to the GitHub Advisory Database entry, Kyverno from 1.16.0 before 1.19.2 does not validate path segments in the resource argument of resource.get() and resource.post() in namespaced CEL policies.
If you run Kyverno on a shared or multi-tenant cluster and let teams create their own policies, treat this as a priority. Clusters where only trusted administrators can create policies are less exposed, but upgrading is still the right fix.
What happened
Kyverno lets teams write validation policies in CEL (Common Expression Language). A NamespacedValidatingPolicy is meant to apply only inside one namespace, which is what makes it reasonable to hand to a tenant. The advisory says the resource.get() and resource.post() functions accept a resource argument whose path segments are not validated. A tenant who can create such a policy can insert .. segments and reach objects outside their namespace.
The advisory states the policy can then read and write arbitrary cluster objects using the admission controller's ServiceAccount, and describes the outcome as cluster takeover. The record was published on November 5, 2026. The NVD entry repeats the same text, so it is not separate confirmation.
Technical details in plain terms
Path traversal (CWE-22) happens when software builds a path from user-supplied input and does not strip out parent-directory markers such as ... Here the path is the one the policy function uses to address a resource. Instead of staying inside the tenant's namespace, the request can climb out of it.
The important part is whose identity is used. The advisory says the access happens through the admission controller's ServiceAccount, and it describes the result as cluster takeover, which implies that account has very wide reach. A low-privileged tenant effectively borrows that identity.
The CVSS v4 vector in the advisory lists network attack vector, low attack complexity, no attack requirements, low privileges required, no user interaction, and high confidentiality, integrity and availability impact on both the vulnerable and subsequent systems.
Who is affected
- Clusters running Kyverno 1.16.0 or later but earlier than 1.19.2.
- Platforms where application teams, customers or CI pipelines hold RBAC permission to create NamespacedValidatingPolicy objects.
- Shared clusters hosting several tenants, where namespace isolation is a security boundary rather than a convenience.
Versions before 1.16.0 are outside the range given in the advisory. If only cluster administrators can create policies, an attacker would first need that access or a compromised tenant account, which lowers but does not remove the risk.
How to check your exposure
The first command lists Deployments across all namespaces and filters for Kyverno, so you can read the image tag and confirm the running version. Run it with kubectl pointed at the cluster you want to check.
kubectl get deployments -A -o wide | grep -i kyvernoIf you installed Kyverno with Helm, this read-only listing shows the chart and app version of each release. Look for an app version from 1.16.0 up to anything below 1.19.2.
helm list -A | grep -i kyvernoNext, confirm the exact policy resource names your cluster knows about. This lists the registered resource types that contain "policies", so you can see whether the namespaced policy type is present.
kubectl api-resources | grep -i policiesThen list existing namespaced validating policies, using the resource name shown by the previous command. Read through them for resource.get() or resource.post() calls whose resource argument contains .. segments. Anything unexpected deserves investigation.
kubectl get namespacedvalidatingpolicies -AWhat to do now
- Upgrade Kyverno to 1.19.2 or later, and confirm the fixed version against the vendor advisory linked above.
- Until you can upgrade, restrict RBAC so only trusted administrators can create or update NamespacedValidatingPolicy objects.
- Review all existing NamespacedValidatingPolicy objects for resource.get() and resource.post() calls with unusual resource arguments, especially ones containing ...
- Review who holds permission to create policies, including service accounts used by CI/CD and third-party tools.
- General practice: check Kubernetes audit logs, if enabled, for policy creation and changes by tenant accounts.
- General practice: if you find a suspicious policy, remove it and consider secrets and objects reachable by the admission controller's ServiceAccount as potentially exposed, then rotate them.
- Re-check the Kyverno version after the upgrade using the commands above.
The steps about audit logs and secret rotation are general incident-response practice, not guidance from the Kyverno advisory. Adapt them to your own environment.
Detection and monitoring
No exploitation has been reported in the sources we reviewed, but a tenant-created policy is a quiet place to hide abuse. As general hardening, consider alerting on creates and updates of policy objects by anyone other than your platform team, where your audit logging allows it.
For broader context on keeping an environment tidy, our small business website security guide covers basics that apply beyond Kubernetes.
If you also run public websites or APIs on the cluster, regular outside-in checks help you notice unexpected changes. Webforta Monitor tracks certificate, DNS and header changes, though it does not see inside your cluster or detect this flaw.
Timeline
| Date | Event |
|---|---|
| 2026-11-05 | CVE-2026-122311 published by the National Vulnerability Database |
| 2026-11-05 | Entry published and last updated in the GitHub Advisory Database; marked unreviewed |
As of this writing only one independent publisher covers the story. Details such as proof of exploitation or additional fixed branches are not stated, so treat anything beyond the advisory text as unconfirmed.
Sources
This article is based on the public sources below. Details can change as vendors and researchers publish more; check the original advisories before acting on version numbers.
