GPU VulnDB

Database/Container, Kubernetes & orchestration

Rancher Fleet: missing valuesFrom validation lets one tenant read another tenant's Fleet credentials

CVSS 9.9CVE-2026-44935Container, Kubernetes & orchestrationcurated

Impact

Fleet is the GitOps controller that many Rancher-managed GPU clusters use to push workloads to downstream clusters. The Helm Deployer does not validate valuesFrom references, so the owner of one Fleet tenant can point a bundle at secrets belonging to another tenant and read them. Those secrets are typically git credentials, registry pulls and cluster-registration tokens, so the payoff is control over another tenant's deployment pipeline rather than a single leaked value. On a shared GPU estate where tenants are separated by Fleet workspaces, this collapses that boundary.

Who can reach it

An authenticated user who owns a Fleet tenant/workspace and can create or edit a bundle - no cluster-admin and no node access needed. Network reach to the Rancher management API is enough.

What to do

Upgrade Fleet to 0.15.2, 0.14.6, 0.13.11 or 0.12.15 depending on your branch. This is a controller upgrade in the management cluster - the Fleet controller pods restart, in-flight bundle reconciles retry, and GPU workloads already running are untouched. No node drain or reboot. Treat any credential reachable via valuesFrom in a multi-tenant workspace as potentially exposed and rotate it.

References

Related entries

All Container, Kubernetes & orchestration 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.