CRI-O EUVD-2026-65156

Exploitation requires three simultaneous conditions: (1) the attacker must have high-privilege access (PR:H) sufficient to submit CreateContainer requests to CRI-O, typically cluster-admin or equivalent pod-scheduling rights in OpenShift; (2) the CreateContainer gRPC request must carry a nil CRI Envs field, triggering the unvalidated fallback path rather than the normal merge path that performs key=value validation; (3) the OCI image referenced in that request must have at least one entry in its config.Env that lacks an '=' character. Exploitation is limited to local access (AV:L) - the CRI-O gRPC socket is not directly network-exposed; the attack flows through the kubelet or equivalent local caller. There is no confidentiality or integrity impact - the only consequence is availability loss. In default OpenShift deployments, the PR:H requirement is a genuine barrier; however, in multi-tenant clusters where tenants control their own image builds, the effective privilege bar may be lower if image content policies are not enforced.

The NVD-assigned CVSS 3.1 score of 6.0 (AV:L/AC:L/PR:H/UI:N/S:C/C:N/I:N/A:H) correctly captures several key risk-limiting factors. The local attack vector (AV:L) reflects that CRI-O's CreateContainer interface is a local gRPC socket consumed by the kubelet rather than a directly network-reachable service. High privilege requirement (PR:H) is the most significant risk limiter: an attacker must have sufficient cluster privileges to schedule workloads using arbitrary OCI images. The scope change (S:C) is the primary risk amplifier - a single crafted container creation crashes the crio daemon for the entire node, killing all unrelated workloads. No EPSS data was provided in the available intelligence, and no KEV listing or POC has been identified, meaning exploitation probability from base rates is unquantifiable but the required privilege level makes opportunistic exploitation unlikely. The realistic threat model is an insider or a compromised account with workload-creation rights deliberately triggering a node outage, rather than external exploitation. In multi-tenant OpenShift clusters where tenants can supply their own images, the privilege bar may be lower than PR:H implies - defenders should verify whether tenant image policies restrict config.Env contents.

An attacker with high cluster privileges (e.g., a compromised service account with pod-creation rights) builds or pushes an OCI image to a reachable registry whose config.Env contains a bare string such as 'NOEQUALS' (no equals sign). They then schedule a pod using that image on a target node, triggering a CreateContainer gRPC call to CRI-O with a nil Envs field; the unvalidated fallback path panics on the malformed env entry, crashing the crio daemon and terminating all container workloads on that node until an operator manually restarts the service. No public exploit code has been identified at time of analysis.