CVE-2026-103002
Description
A flaw was found in GPUStack. An authenticated remote user without administrator privileges can forward requests through the cluster API server proxy to a connected Kubernetes cluster without proper resource-level authorization checks. Because the proxy executes requests using the credentials of a worker node service account, an attacker can read sensitive resources, manipulate workloads, and potentially achieve arbitrary code execution within the target cluster.
Statement
This vulnerability is rated as Important rather than Critical because exploitation requires valid user authentication and a reachable worker node, though successful attacks permit complete cluster workload manipulation and code execution. GPUStack is not included or supported in Red Hat Enterprise Linux or Red Hat OpenShift, limiting direct exposure on default platform deployments. In environments where third-party GPUStack instances manage or proxy requests to Red Hat OpenShift clusters, the lack of resource authorization enables non-administrative users to abuse worker ServiceAccount privileges.
Mitigation
Mitigation for this issue is either not available or the currently available options do not meet the Red Hat Product Security criteria comprising ease of use and deployment, applicability to widespread installation base, or stability.
--- CME Mitigation Guidance (2026-11-06) ---
Disclaimer: The following mitigation solutions are provided as a courtesy, pending the application of an official patch. This content was generated utilizing AI tools in conjunction with data from https://cmetaxonomy.org. Customers are advised to independently evaluate and test these temporary mitigations within their own infrastructure.
The highest-value action is to enforce tenant scoping and resource-level authorization checks on the GPUStack cluster API server proxy before forwarding requests using worker node ServiceAccount credentials: upgrade to GPUStack 2.2.2, which resolve caller tenant context, require write permissions for manifests, and restrict non-manager proxy requests to read-only workload APIs in their principal namespace, closing all 2 injection vectors (arbitrary Kubernetes API request forwarding via cluster proxy, unauthorized cluster manifest retrieval). Until patched, this is a concern on GPUStack deployments managing Kubernetes-provider clusters with reachable worker nodes, so Restrict GPUStack access strictly to trusted administrator accounts and audit non-administrative user permissions on clusters with reachable worker nodes. As a compensating measure, Enable Kubernetes API server audit logging to detect privileged resource access, pod creation, and exec requests forwarded using worker ServiceAccount tokens, Enforce restricted Pod Security Standards in the target Kubernetes cluster to limit capabilities of workloads created via abused worker credentials, Deploy admission controller policies with OPA, Gatekeeper, or Kyverno to intercept and reject unauthorized or privileged pod manifests, Enforce URL path authorization on cluster proxy and manifest forwarding endpoints based on caller privileges before reaching proxy logic, and Deploy Falco or eBPF runtime security monitoring to detect anomalous container launches, sensitive file reads, and shell spawns. These controls limit blast radius and contain exploitation, but they do not fix the underlying missing resource-level authorization defect — apply the upstream patch to fully remediate.
Highest-impact CME: CME-907 Application-Layer RBAC Enforcement (high)
Understanding the Weakness (CWE)
Confidentiality,Integrity,Availability,Access Control
Technical Impact: Read Application Data; Modify Application Data; Execute Unauthorized Code or Commands; Gain Privileges or Assume Identity
Frequently Asked Questions
Not sure what something means? Check out our Security Glossary.
Want to get errata notifications? Sign up here.
