On this page (6 sections)
A flaw tracked as CVE-2026-122321 affects the image verification feature of Kyverno, the Kubernetes policy engine. In Kyverno 1.18.0 before 1.19.2, a cache used by verifyImages can cause signature and attestation checks to be skipped for an image that was never actually verified.
If you run Kyverno to enforce signed container images, especially on clusters where several teams or tenants can submit workloads, you should check your version now. At the time of writing, the details come from a single advisory chain, so treat the specifics below as what that advisory states.
What happened
According to the NVD record and the GitHub advisory, Kyverno's verifyImages feature caches the result of a successful verification. The cache is keyed by the image reference string, such as a tag, and not bound to the specific digest that was verified. When a later request hits the cache, Kyverno skips the signature and attestation checks.
The advisory was published on November 5, 2026. The NVD entry repeats the GitHub text, so there is one underlying description rather than two independent confirmations. A VulnCheck advisory and a Kyverno project advisory are listed as references, but their content was not part of the research for this article.
Why a tag-keyed cache is a problem
A tag such as app:v1 is a movable label. The registry owner, or anyone with push rights to that repository, can point it at a different image at any time. A digest, by contrast, identifies one exact image content. Signature checks are only meaningful when they apply to a digest.
The weakness class here is a time-of-check time-of-use problem (CWE-367): the thing that was checked (a signed image behind a tag) is not guaranteed to be the thing that is used later (whatever the tag points to by then). The cache makes that gap persist.
The scenario described in the advisory
The advisory describes an attacker who can submit workloads and can move a registry tag. They first get a signed image admitted, which fills the cache for that tag. They then repoint the tag to a different image. On the next admission, the cache hit skips verification, Kyverno admits the workload, and its mutateDigest behavior pins the digest of the unverified image. The result is a workload that looks policy-compliant and digest-pinned but was never verified. No exploit code is needed to understand the risk, and none is given here.
Who is affected
Exposure requires several conditions to line up, based on the advisory text:
- You run Kyverno 1.18.0 up to, but not including, 1.19.2.
- You use verifyImages rules to enforce image signatures or attestations.
- An attacker can submit workloads to the cluster (low privileges are enough per the CVSS metrics).
- The attacker can also change what a tag points to in a registry the cluster pulls from.
The CVSS v4 vector lists network attack vector, low complexity, no attack requirements, low privileges, no user interaction, and high integrity impact. Multi-tenant clusters, shared CI namespaces and setups that pull from registries where many people can push are the most realistic exposure. A single-team cluster pulling from a locked-down registry has a much smaller attack surface, though it should still be updated.
How to check your exposure
Start by finding which Kyverno version is running. The first command lists the container images used in the kyverno namespace; look for a tag between 1.18.0 and 1.19.1. Adjust the namespace if you installed Kyverno elsewhere.
kubectl get pods -n kyverno -o jsonpath='{range .items[*]}{.metadata.name}{"\t"}{range .spec.containers[*]}{.image}{" "}{end}{"\n"}{end}'Next, see whether you actually use image verification. This read-only command lists cluster-wide policies so you can review which contain verifyImages rules. Repeat with the namespaced policy resource if you use it.
kubectl get clusterpoliciesIf you installed Kyverno with Helm, this read-only command shows the installed release, chart and app version in the kyverno namespace.
helm list -n kyvernoWhat to do now
The affected range in the CVE description (1.18.0 before 1.19.2) indicates that 1.19.2 is the first release without the flaw. The GitHub entry lists patched versions as unknown, so confirm the fixed version in the Kyverno release notes. Upgrade through the same method you used to install Kyverno, and test in a non-production cluster first, since admission controller changes can affect every workload deployment.
- Confirm the Kyverno version in every cluster, including staging and CI clusters.
- Upgrade any 1.18.0 to 1.19.1 installation to 1.19.2 or later, after confirming the fixed version in the Kyverno release notes.
- Review which policies use verifyImages and which namespaces they cover.
- Limit who can push to, or retag, the registries your cluster pulls from.
- Restrict who can create workloads in namespaces covered by signature policies.
- Prefer referencing images by digest in your own manifests so a moved tag cannot change what runs.
- Audit recent workloads in sensitive namespaces and compare running image digests with what your build pipeline produced.
kubectl get pods -n NAMESPACE -o jsonpath='{range .items[*]}{.metadata.name}{"\t"}{range .status.containerStatuses[*]}{.imageID}{" "}{end}{"\n"}{end}'Detection and monitoring
The research does not describe any in-the-wild exploitation or specific indicators, so there is no known signature to hunt for. Practical monitoring is about verifying outcomes: alert on workloads whose running digest does not match a signed build, on tag changes in your registry for repositories that feed production, and on new workloads appearing in namespaces protected by signature policies.
For the public websites your clusters serve, outside-in monitoring of certificates, DNS and security headers is a separate layer; Webforta Monitor covers that. It does not look inside your cluster and does not address this flaw.
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.
