Database/Container, Kubernetes & orchestration
aws-privateca-issuer: namespace-scoped issuers fall back to the controller's own AWS identity
Impact
A namespace-scoped issuer that supplies no credentials, or sets only a role, uses the controller's own AWS identity. A user who can create such an issuer in one namespace can then get certificates, including subordinate CA certificates, from every AWS Private CA and certificate template that the controller identity can reach, or reach through any IAM role it can assume. On a multi-tenant GPU cluster where tenants own namespaces, one tenant can obtain certificates that other tenants' services or cluster infrastructure trust. A subordinate CA certificate would let them issue further certificates of their own. The flaw is in the credential handling, so the scope of exposure is set by what the controller's IAM identity is allowed to do.
Who can reach it
An authenticated Kubernetes user with RBAC permission to create namespace-scoped issuers (AWSPCAIssuer) in any namespace. Network reachability to the API server is needed; no AWS credentials of their own are needed.
What to do
Before upgrading, change every namespace-scoped issuer that relies on the controller's identity so that it supplies its own credentials. Then upgrade aws-privateca-issuer to 1.12.0, or the EKS add-on to v1.12.0-eksbuild.1. The upgrade rolls the controller pod and does not touch the nodes. Until then, restrict who can create namespace-scoped issuers and narrow the controller's IAM permissions and assumable roles. Review certificates issued from your Private CAs, especially subordinate CA certificates, for unexpected requesters.
References
Related entries
- OpenRun: redirect URL restrictions can be bypassed, giving an open redirect on the deploy consoleCVE-2026-55252 · OpenRun (self-hosted GitOps platform, redirect URL validation)Medium
- AWS EFS CSI Driver: crafted volumeHandle causes recursive deletion of directories on another filesystemCVE-2026-85781 · Amazon EFS CSI Driver (volume deletion, access point ownership check)Medium
- CRI-O: All pod processes share one memory cgroup, so a workload OOM kills conmon and destabilises the nodeCVE-2019-14891 · CRI-OMedium
- containerd: Crafted image can change Unix file permissions of existing host files during extractionCVE-2021-32760 · containerdMedium
- Istio: A user with CREATE on Gateway API resources escalates privilege in istiodCVE-2022-21701 · IstioMedium
- runc: Rootless runc leaves /sys/fs/cgroup writable inside the containerCVE-2023-25809 · runcMedium
This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.