GPU VulnDB

Database/Container, Kubernetes & orchestration

Rancher: plaintext non-expiring registration tokens let an attacker join a rogue node to a cluster

CVE-2026-55997Container, Kubernetes & orchestrationcurated

Impact

Registration tokens that authenticate nodes and agents joining a downstream cluster were stored in plaintext and never expired, so any copy obtained once is valid forever. An attacker who reads one - via the Rancher API, etcd, a backup, stored automation, or a file on an existing node - can register a machine they control into the cluster and start receiving scheduled work. On a GPU fleet that means tenant training jobs, mounted datasets and model weights can be dispatched to attacker hardware, and the rogue node gets a legitimate kubelet identity inside the cluster. Rotation is the expensive part, because these tokens are typically baked into node bootstrap automation and golden images.

Who can reach it

Anyone who can read a registration token: an authenticated Rancher user with basic API access, anyone holding etcd or backup access, or a local user on an existing cluster node. Registering the rogue node afterwards needs no further credential.

What to do

Upgrade Rancher to the version named in the vendor advisory - the record does not state a fixed version, so read GHSA-7r53-jvhg-9jq4 before scheduling. Upgrading the Rancher management plane is a controller restart, not a fleet event: GPU nodes keep running. The work that follows the patch is the real cost - rotate every existing registration token, then update the bootstrap automation and node images that embed the old ones, since already-leaked tokens stay valid until rotated. Audit recently joined nodes for machines you did not provision.

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.