Three security vulnerabilities have been discovered in the widely used container runtime containerd, putting Kubernetes environments at risk. The most severe, CVE-2026-5348 (CVSS 9.4), allows arbitrary command execution on the host. Containerd is used by Kubernetes, Google GKE, Amazon EKS, Azure AKS, and internally by Docker.
CVE-2026-5348: Arbitrary Host Command Execution
The CRI plugin, which handles communication between Kubernetes and containerd, transferred image labels to containers without validation in versions before 1.7.33, 2.3.2, 2.2.5, 2.1.9, and 2.0.10. When plugins use these labels for certain operations, an attacker can leverage them to run arbitrary commands on the host. The fix is included in the versions listed above.
CVE-2026-53492: Resource Control Bypass
CVE-2026-53492 (CVSS 8.4) is also worth attention. During container restore from a checkpoint, containerd pulls CDI annotations from the checkpoint archive rather than the original pod specification. A user with pod creation permissions can exploit this to bypass Kubernetes resource allocation and device plugin constraints. CDI must be enabled on the affected node. Versions 2.1.0 through 2.1.8, 2.2.0 through 2.2.4, and 2.3.0 through 2.3.1 are affected.
CVE-2026-53489: Access to Arbitrary Host Files
CVE-2026-53489 (CVSS 8.2) also involves container restore. A crafted symbolic link for container.log causes kubectl logs to read an arbitrary host file instead of the container log, potentially exposing configuration files or credentials. All three vulnerabilities are resolved by updating to the patched containerd versions.