CVE-2026-104623: 
Analyse et atténuation de la vulnérabilité

CVE-2026-104623 is an authorization bypass vulnerability in Rancher Manager's local (management) cluster shell endpoint that allows any authenticated Rancher user to obtain an interactive shell pod on the management cluster without requiring any cluster-level permissions. The vulnerability affects Rancher versions >=2.11.0 <2.11.19, >=2.12.0 <2.12.15, >=2.13.0 <2.13.11, >=2.14.0 <2.14.7, and >=2.15.0 <2.15.3. It was published on October 29, 2026, and carries a CVSS v4 base score of 5.3 (Moderate) (Rancher Advisory).

The root cause is a missing/incorrect authorization check (CWE-862, CWE-863) on the local cluster shell API handler. The authorization gate introduced to protect this feature (as a fix for CVE-2022-21953) was only applied to one specific API route form, leaving alternative request forms able to reach the shell handler without triggering the access check. An attacker with only a default Rancher user account (no local cluster role bindings) can craft an HTTP request using an alternative route form to bypass the authorization check and spawn a dashboard-shell-* pod in the cattle-system namespace on the management cluster (Rancher Advisory).

Successful exploitation allows any authenticated Rancher user to obtain an interactive shell pod running in the cattle-system namespace on the management cluster, breaking the expected isolation of the management plane. The spawned pod runs unprivileged and the user's Kubernetes API access remains scoped to their own identity, so no additional Kubernetes RBAC permissions are directly granted. However, the attacker gains unauthorized compute resources and an in-namespace network position on the management cluster, which could facilitate reconnaissance, lateral movement within the namespace, or abuse of network adjacency to other management-plane workloads (Rancher Advisory).

Exploitation requires only a valid authenticated Rancher account — no elevated privileges, no local cluster role bindings, and no non-default feature flags are needed. Default-configuration deployments are affected. No public PoC code or in-the-wild exploitation has been reported in available sources, and no CISA KEV listing or EPSS score data is currently available for this CVE. The vulnerability was discovered and reported by Kenzie Cuttler (@K-Cuttler) and Arpit Jain (@arpitjain099) (Rancher Advisory).

1. Obtain a Rancher account: Acquire any authenticated Rancher user account — even a default user with no cluster-level permissions is sufficient.
2. Identify the target: Confirm the Rancher instance is running an affected version (>=2.11.0 <2.11.19, >=2.12.0 <2.12.15, >=2.13.0 <2.13.11, >=2.14.0 <2.14.7, or >=2.15.0 <2.15.3).
3. Craft an alternative API request: Instead of using the primary API route for the local cluster shell (which is protected by the authorization check), send an HTTP request using an alternative route form that reaches the shell handler without triggering the access gate.
4. Spawn a shell pod: The request causes Rancher to create a dashboard-shell-* pod in the cattle-system namespace on the management cluster.
5. Obtain interactive shell: Connect to the spawned pod to gain an interactive shell on the management cluster, with network adjacency to other management-plane workloads (Rancher Advisory).

• Kubernetes Resources: Unexpected dashboard-shell-* pods in the cattle-system namespace on the local (management) cluster not associated with legitimate administrator activity.
• Logs: Rancher audit logs showing shell requests to the local cluster shell endpoint from users who do not hold role bindings on the local cluster or administrative privileges.
• Network: Unexpected interactive sessions or network connections originating from dashboard-shell-* pods within the cattle-system namespace (Rancher Advisory).

Upgrade to a patched version of Rancher: v2.15.3, v2.14.7, v2.13.11, v2.12.15, or v2.11.19. No workaround fully addresses the issue, but organizations unable to upgrade immediately can reduce exposure by: restricting the set of users holding Rancher accounts; placing Rancher behind a reverse proxy, ingress policy, or WAF that blocks requests to the local cluster shell endpoints from all but trusted administrative networks; and monitoring the cattle-system namespace for unexpected dashboard-shell-* pods. Administrators should also review Rancher audit logs for unexpected shell requests and remove any unauthorized pods found (Rancher Advisory).