AWS identified five security issues in the containerd CRI plugin versions 1.7 through 2.3: CVE-2026-50195 allows image cache poisoning via unvalidated checkpoint image references enabling cross-pod code execution; CVE-2026-53488 enables arbitrary host command execution through unsanitized image LABEL instructions; CVE-2026-53492 permits device and host mount injection by trusting unvalidated CDI annotations; CVE-2026-53489 allows arbitrary host file read via unvalidated symlinked container log paths during checkpoint restore; CVE-2026-47262 causes denial of service by triggering uncontrolled memory consumption from crafted container images. These issues affect containerd runtimes underpinning AWS managed container services and self-managed deployments. Patched versions are available upstream, and AWS is rolling out fixes for managed services.

The vulnerabilities allow for cross-pod code execution, arbitrary host command execution, device and host mount injection bypassing Kubernetes enforcement, arbitrary host file read, and denial of service through containerd process termination. These impacts can compromise container isolation, lead to unauthorized access to host resources, and disrupt container workloads. The severity of individual issues ranges from medium to high based on CVSS scores provided by AWS (6.5 to 8.8).

Patched containerd releases addressing these vulnerabilities are available upstream and should be applied promptly. AWS is deploying patched runtimes across affected managed container service fleets. For self-managed containerd deployments, upgrade to a patched version immediately. As workarounds, disabling the checkpoint/restore feature mitigates CVE-2026-50195, CVE-2026-53489, and CVE-2026-53492; additionally, disabling CDI support mitigates CVE-2026-53492. There are no workarounds for CVE-2026-53488 and CVE-2026-47262 other than upgrading.