2026-10-26
Cluster add-ons stop trusting the compromised eMudhra CA certificates
#add-on #kubernetes #security #update #upgrade #component #eks
Last week, the private keys of eight eMudhra intermediate certificate authorities were found to be compromised (CVE-2026-111943). Anyone holding them can issue TLS certificates that look valid. Browsers and operating systems are removing eMudhra from their trusted certificates, and the projects behind our cluster add-ons are shipping rebuilt releases. We are rolling out a follow-up to the October add-on upgrades that picks them up. It is live on non-production clusters, and production follows.
What changed
- Add-ons that make outgoing TLS connections no longer trust eMudhra. cert-manager, the AWS Load Balancer Controller, Dex, the blackbox exporter and the EBS CSI driver now run images without the compromised eMudhra certificates. This matters most for cert-manager, which talks to your ACME issuer and DNS provider, and for Dex, which talks to your login provider.
- Other add-ons reject the compromised certificates as well. Alloy, Loki, Tempo, the Vertical Pod Autoscaler, Karpenter and the EKS node monitoring agent are rebuilt with a Go version that refuses certificates signed by the compromised intermediates, even when they are still in the image’s trust store.
- A new node image. EKS node image v20261023 also blocks the compromised certificates. It replaces every node once, and Karpenter moves your workloads within your node pools’ disruption budgets, as in any node rotation.
- Traefik 3.7.16 closes four security issues. If you route traffic with your own Gateway API
HTTPRoutes, a backend on port 443 or on a port namedhttpsnow needsappProtocol: httpson its Service port to keep using HTTPS. Ingresses andIngressRoutes are not affected. - GPU node checks are stricter. The EKS node monitoring agent now also checks the NVIDIA driver and Fabric Manager. On GPU clusters, a node with a hung driver is now flagged as unhealthy and replaced, where it used to keep running.
What you need to do
Nothing for the add-ons we manage. Your own applications are a different story: an image built before last week most likely still trusts the eMudhra certificates. Rebuild your images on an up-to-date base image (Alpine ca-certificates 20261020 or newer, or a recent distroless or Debian image), and move Go services to Go 1.26.10 or 1.27.3 or newer.
Add-on upgrades
- alloy v1.21.2 (chart v1.14.3)
- aws-ebs-csi-driver v1.67.2
- Fixes the eMudhra CVE; see above.
- aws-efs-csi-driver v3.5.2
- aws-load-balancer-controller v3.6.2
- Image without the eMudhra certificates.
- cert-manager v1.21.3
- Image without the eMudhra certificates.
- dex v2.46.1 (chart v0.26.0)
- Image without the eMudhra certificates.
- eks-node-monitoring-agent v1.7.5
- Stricter GPU node checks; see above.
- karpenter v1.15.1
- kube-prometheus-stack v92.13.1 (Prometheus Operator v0.94.1)
- grafana-loki v3.7.9 (chart v20.0.1)
- Major chart version; your logs and log retention are not affected.
- prometheus-blackbox-exporter v0.29.1 (chart v11.20.2)
- Image without the eMudhra certificates.
- snapshot-controller v8.7.0
- tempo v3.1.1 (chart v3.2.1)
- traefik v3.7.16 (chart v42.1.3)
- Security fixes and a Gateway API change; see above.
- vertical-pod-autoscaler v1.8.1 (chart v0.15.0)
- wg-easy v15.5.0
We also picked up the latest EKS build revisions for CoreDNS, kube-proxy, metrics-server and the VPC CNI (packaging updates only, no version change).
Some add-ons have no release yet that drops the eMudhra certificates, among them Flux, KEDA, external-dns, fluent-bit and Velero. We upgrade them as soon as their maintainers publish one.