GPU VulnDB

Database/Container, Kubernetes & orchestration

Chaos Mesh Chaos Controller Manager: unauthenticated GraphQL debug server kills processes cluster-wide

CVSS 7.5CVE-2025-59358Container, Kubernetes & orchestrationcurated

Impact

The Chaos Controller Manager listens for GraphQL debug queries with no authentication and is reachable from anywhere inside the cluster pod network. The schema exposes a mutation that kills arbitrary processes in any pod on the cluster, so one compromised or hostile workload can terminate neighbours' processes at will. On a GPU cluster that means a tenant pod can kill training ranks, the device plugin, DCGM exporter or node-level agents belonging to other tenants, and long multi-node jobs lose hours of work with no checkpoint. The reported impact is denial of service only - availability, not confidentiality or integrity - but Chaos Mesh is usually installed with broad cluster privileges, so this is a tenancy boundary that exists on paper and not in the code. Only install it on clusters where every workload is trusted, or where the controller is network-policy isolated.

Who can reach it

Any workload that can open a TCP connection to the Chaos Controller Manager service inside the Kubernetes cluster - any tenant pod on a shared cluster. No authentication and no Kubernetes RBAC permission is required.

What to do

Upgrade Chaos Mesh to the release carrying the fix in chaos-mesh PR #4702, which gates the GraphQL debug server; this is a controller Deployment rollout, no node drain or reboot. If you cannot upgrade immediately, disable the debug server in the Helm values and apply a NetworkPolicy that allows only the Chaos Mesh namespace to reach the controller, or uninstall Chaos Mesh from clusters that do not currently need it. The advisory does not state a fixed version number here, so check the chaos-mesh release notes before scheduling.

References

https://nvd.nist.gov/vuln/detail/CVE-2025-59358https://github.com/chaos-mesh/chaos-mesh/pull/4702https://jfrog.com/blog/chaotic-deputy-critical-vulnerabilities-in-chaos-mesh-lead-to-kubernetes-cluster-takeover

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.