On this page (7 sections)
A critical vulnerability, CVE-2026-122310, affects Kyverno from version 1.19.1 up to but not including 1.19.2. According to the GitHub Advisory Database entry, a namespace-scoped user who is allowed to create a Policy can make Kyverno call cluster-scoped API endpoints using Kyverno's own ServiceAccount, which ends in cluster admin.
If you run Kyverno as an admission controller and let teams create namespaced Policies, this is relevant to you. Platform engineers, hosting administrators and agencies running shared Kubernetes clusters should check their Kyverno version now.
What happened
The advisory, published on November 5, 2026, describes a mismatch between what Kyverno checks and what it runs. Kyverno lets a Policy define a context entry with an apiCall, which fetches data from the Kubernetes API through a urlPath. In the affected versions, Kyverno validates that path after cleaning it, but then executes the raw, uncleaned path.
The Kubernetes client discards any fragment (after #) or query (after ?) in that raw path. According to the advisory, that gap lets a user hide ../ segments behind a # or ? so the validated path looks harmless while the request that is actually sent points at a cluster-scoped endpoint.
Why it matters: the weakness in plain terms
This is a validation-versus-use problem, classified as CWE-436. Two components interpret the same string differently: the checker sees one thing, the executor does another. Security checks are only meaningful if they run on exactly the value that is later used.
The impact comes from whose identity makes the request. Per the advisory, the request is made with Kyverno's ServiceAccount, so a user limited to one namespace can reach cluster-scoped endpoints with that identity. The advisory states the result is cluster admin.
The CVSS v4 metrics in the GitHub entry show 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 realistically exposed
- Clusters running Kyverno 1.19.1, the only version in the stated affected range before 1.19.2.
- Multi-tenant clusters where developers or tenants have rights to create Policy objects in their own namespaces.
- Environments where a compromised developer account or CI token could create namespaced Policies.
- Single-admin clusters where nobody but cluster admins can create Policies carry lower practical risk, since the attacker already needs that privilege, but should still update.
The exploit requires the ability to create a Policy, so it is a privilege escalation rather than an unauthenticated remote attack. Treat any account with that right as a path to full cluster control until you have patched.
How to check your exposure
First find out which Kyverno version is running. This read-only command lists container images in the kyverno namespace; look for a tag between 1.19.1 and below 1.19.2. The namespace name kyverno is the common default, so adjust it if yours differs.
kubectl get pods -n kyverno -o jsonpath='{range .items[*]}{.metadata.name}{"\t"}{range .spec.containers[*]}{.image}{" "}{end}{"\n"}{end}'Next, see who can create Policies. This read-only check shows whether a given user may create namespaced Policies in a namespace. Replace the user and namespace with real values; a result of yes means that identity is within the scope of this issue.
kubectl auth can-i create policies.kyverno.io --as=<user> -n <namespace>Finally, review existing namespaced Policies that use context apiCall entries, especially ones you did not write. This read-only command prints lines mentioning urlPath or apiCall across all namespaces; inspect the urlPath values it finds for unexpected # or ? characters or ../ sequences.
kubectl get policies.kyverno.io -A -o yaml | grep -n -i -E 'urlPath|apiCall'What to do now
The fix is to move to Kyverno 1.19.2 or later. The research does not list a workaround, so patching is the supported route.
- Identify every cluster running Kyverno and record the version.
- Upgrade any 1.19.1 installation to 1.19.2 or later using your usual Helm or manifest process.
- Restrict who can create Policy objects in namespaces until the upgrade is done.
- Review existing namespaced Policies with apiCall context entries for suspicious urlPath values.
- Review audit logs for Policy creation by non-admin users and for unexpected cluster-scoped reads from Kyverno's ServiceAccount.
- If you find signs of abuse, treat the cluster as compromised: rotate credentials and review RBAC bindings.
After the upgrade, confirm the new version with the same image listing command used above.
Detection and ongoing monitoring
The research gives no known indicators of compromise, so rely on general Kubernetes hygiene. Enable API server audit logging and alert on creation or modification of Policy objects, on new ClusterRoleBindings, and on unusual requests made by the Kyverno ServiceAccount to cluster-scoped paths.
Apply least privilege to Policy authoring. Few tenants truly need to create Policies, and removing that right shrinks the audience for this class of bug. If you want a wider view of your public-facing surface, a read-only Webforta scan covers externally visible issues, though it cannot see inside your cluster's RBAC.
Source caveats
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.
