containerd CVE-2026-47262

Exploitation requires that a node running an unpatched containerd version creates a container from the attacker's maliciously crafted image - standard container creation via any interface (kubectl, Docker CLI, CRI) is sufficient, with no non-default feature (such as CDI, checkpoint/restore, or CRI plugins) required. The primary limiting factor is image delivery: the attacker must place the malicious image in a registry accessible to the target node, and an operator or automated process must then instantiate a container from it. Environments enforcing strict image allowlists via admission control, content-trust verification (Sigstore/Cosign), or private registries with controlled push access are significantly harder to exploit. Impact is strictly limited to availability - no code execution, privilege escalation, or data exfiltration is described. No special configuration is required beyond having containerd run an attacker-supplied image.

No CVSS vector or EPSS score is provided in the available intelligence for CVE-2026-47262. The containerd project's own Moderate rating situates this below the three Critical CVEs (CVE-2026-50195, CVE-2026-53488, CVE-2026-53492) and one High CVE (CVE-2026-53489) patched in the same release batch - all of which involve code execution or privilege escalation rather than availability alone. The DoS impact is transient in principle: restarting containerd restores runtime functionality. However, in production Kubernetes clusters, an OOM kill of containerd causes the node to transition to NotReady, may trigger mass pod eviction, and can cascade to capacity-sensitive workloads if multiple nodes are simultaneously targeted. Nodes with constrained memory or high pod density are at greater risk of amplified operational impact. The absence of a confidentiality or integrity impact constrains real-world severity, but availability-critical environments (financial services, healthcare, high-throughput SaaS) should treat this as a material operational risk. CVSS vector, EPSS, KEV status, and POC availability are all unconfirmed from available data - risk tier is based solely on the vendor-assigned severity class and the described impact.

An attacker crafts a container image designed to trigger unbounded memory allocation within containerd's parsing logic and hosts it on a public registry or a compromised internal one. A cluster operator, developer, or automated CI/CD pipeline pulls and instantiates the image on a target node running an unpatched containerd version; containerd exhausts available node memory, is killed by the Linux OOM killer, and the container runtime API becomes unavailable - potentially causing the Kubernetes node to enter NotReady state and disrupting all workloads on that node. No public proof-of-concept exploit has been identified at time of analysis.