GPU VulnDB

Database/Control plane, storage & DevOps

GPUStack: cluster API proxy forwards non-admin requests with a worker's ServiceAccount credentials

CVSS 8.8CVE-2026-103002Control plane, storage & DevOpscurated

Impact

In GPUStack 2.2.0 through 2.2.1 the cluster_apiserver_proxy forwarded an authenticated non-administrator's requests to a selected Kubernetes cluster with no resource-level authorization check, using the credentials of a reachable worker's Kubernetes ServiceAccount. Any logged-in GPUStack user therefore acts with that ServiceAccount's full rights: reading secrets and other sensitive resources, creating or modifying workloads, or deleting them. Because the attacker can create workloads, this becomes code execution inside the target cluster - on a GPU cluster manager that means a low-privileged user can schedule their own pods onto GPU nodes and reach other tenants' model-serving workloads. Requires a Kubernetes-provider cluster with a reachable worker.

Who can reach it

Any authenticated GPUStack user who is not an administrator, over the GPUStack API, against a Kubernetes-provider cluster that has a reachable worker.

What to do

Upgrade GPUStack to 2.2.2 or later and restart the GPUStack server. No node reboot. Afterwards, rotate the worker ServiceAccount tokens that the proxy was using and audit the target clusters for workloads created by unexpected principals; tightening the ServiceAccount's RBAC limits the blast radius of any similar gap.

References

Related entries

All Control plane, storage & DevOps entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.