CVE-2026-2270: 
NixOS Analyse et atténuation de la vulnérabilité

CVE-2026-2270 is a control-plane flaw in Kubernetes that allows a pod to be created outside its intended namespace, potentially bypassing user permissions and multi-tenant isolation boundaries. The vulnerability is associated with StatefulSet and ControllerRevision write permissions enabling cross-namespace pod creation via the kube-controller-manager. It affects the Kubernetes project (specific version ranges not yet publicly confirmed, as the CVE remains in Reserved status). The estimated CVSS severity is HIGH (Feedly, OSS-Sec).

The root cause is insufficient namespace validation in the Kubernetes control plane, specifically within the kube-controller-manager when processing StatefulSet and ControllerRevision resources (CWE-284: Improper Access Control). An attacker with write permissions to StatefulSet or ControllerRevision objects can manipulate these resources in a way that causes the controller manager to create pods in a namespace other than the one the user is authorized to access. This is a cluster-wide controller issue where the controller fails to validate that actions performed on behalf of a user remain within the user's permitted namespace scope. Technical analysis has been published describing how ControllerRevision manipulations break multi-tenant isolation (sh3llc0d3.com, Kubernetes Issue).

Successful exploitation allows a user with limited namespace-scoped permissions to create pods in namespaces they are not authorized to access, breaking multi-tenant isolation in shared Kubernetes clusters. This could lead to unauthorized workload execution in sensitive namespaces, potential access to secrets or service account tokens in those namespaces, and lateral movement across tenant boundaries. The vulnerability does not grant access to all namespaces universally, but the specific permission combinations that enable it make it a significant risk in multi-tenant environments (Feedly, dev.to).

As of the time of reporting, no public proof-of-concept exploit code has been confirmed, and there is no evidence of in-the-wild exploitation (Feedly). The CVE remains in Reserved status, meaning full technical details have not been officially published. No EPSS score or CISA KEV catalog entry has been identified for this CVE. Exploitation requires an attacker to already have write access to StatefulSet or ControllerRevision resources within a namespace, limiting the attack surface to authenticated users with specific RBAC permissions (OSS-Sec).

1. Reconnaissance: Identify a Kubernetes cluster running a vulnerable control-plane version. Enumerate available RBAC permissions using kubectl auth can-i --list to confirm write access to StatefulSet or ControllerRevision resources in at least one namespace.
2. Craft malicious ControllerRevision or StatefulSet: Create or modify a ControllerRevision or StatefulSet object with crafted metadata or spec fields that reference a target namespace different from the attacker's authorized namespace.
3. Trigger controller reconciliation: Submit the crafted resource to the API server. The kube-controller-manager processes the object and, due to insufficient namespace validation, initiates pod creation in the target namespace on behalf of the attacker.
4. Achieve cross-namespace pod execution: The pod is created in the target namespace, potentially with access to that namespace's secrets, ConfigMaps, and service account tokens, enabling further lateral movement or privilege escalation within the cluster (sh3llc0d3.com, dev.to).

• Logs: Kubernetes API server audit logs showing a user creating or patching StatefulSet or ControllerRevision objects in one namespace while pods appear in a different namespace shortly after; kube-controller-manager logs referencing pod creation events in unexpected namespaces.
• Kubernetes Events: Pod creation events in namespaces where the initiating user has no direct pod creation permissions, especially correlated with StatefulSet or ControllerRevision modifications in another namespace.
• RBAC Anomalies: Users or service accounts with update/patch rights on statefulsets or controllerrevisions in one namespace but no explicit pod creation rights in the namespace where new pods appear.
• Resource Artifacts: Unexpected pods running in sensitive or system namespaces with owner references pointing to StatefulSets or ControllerRevisions in a different namespace (sh3llc0d3.com).

Patches are available for this vulnerability; administrators should update the Kubernetes control plane to the fixed versions as soon as they are officially published (Feedly). As an interim measure, audit and restrict RBAC permissions — particularly update and patch rights on statefulsets and controllerrevisions — to only trusted users and service accounts. Cluster administrators should review permission combinations that could enable this attack and apply the principle of least privilege. Enabling admission controllers such as OPA/Gatekeeper or Kyverno with policies enforcing namespace boundaries can provide additional defense-in-depth (OSS-Sec, Kubernetes Issue).

The vulnerability was disclosed via the Kubernetes security advisory forum and the OSS-Security mailing list, generating community discussion about multi-tenant isolation risks in Kubernetes (Kubernetes Advisory, OSS-Sec). Security researchers published technical breakdowns highlighting how ControllerRevision manipulation can undermine namespace isolation, and the issue was covered in the "Last Week in Kubernetes" community newsletter (LWKD). The broader community reaction emphasized the need for careful RBAC auditing and the importance of keeping control-plane components updated in shared cluster environments (dev.to).