GPU VulnDB

Database/Container, Kubernetes & orchestration

RHACM app-subscription: namespace edit rights escalate to cluster-admin via attacker-hosted Helm chart

CVE-2026-10090Container, Kubernetes & orchestrationcurated

Impact

A user holding only namespace-scoped edit rights on an ACM hub can create a Channel pointing at a Helm repository they control plus a Subscription referencing it. The subscription controller fetches that chart and applies it with its own elevated authority, without checking for the subscription-admin role and without confining the applied objects to the subscription namespace. Cluster-scoped resources in the chart therefore land as-is, so a ClusterRoleBinding granting the attacker's ServiceAccount cluster-admin gets created for them. On a fleet where the ACM hub governs GPU clusters, that is full control of the hub and, through it, the managed spokes: any tenant workload, node label, taint or scheduling policy on the accelerator fleet becomes editable.

Who can reach it

Authenticated user with edit privileges in any namespace on the ACM hub, plus the ability to host a Helm repository the hub can reach. No prior cluster-scoped permission required.

What to do

Apply the Red Hat errata for your ACM stream (RHSA-2026:60386 through 60390 cover 2.11 and 2.13 through 2.17). This is a hub-side operator update - the ACM operator redeploys the subscription controller pods; managed spoke clusters and GPU nodes are not rebooted. Before patching, audit existing Channel and Subscription objects in tenant namespaces and review ClusterRoleBindings created since the affected controllers were deployed, since exploitation leaves a durable binding behind.

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.