Database/AI/ML frameworks & serving
NVIDIA OpenShell: plaintext SSH handshake secret in sandbox environment lets one sandbox reach others
Impact
In OpenShell's Kubernetes deployment, the SSH handshake secret sits in a plaintext environment variable inside the sandbox. Because the secret is shared across sandboxes, a low-privileged process in one sandbox can read it and use it to reach other sandboxes. On a multi-tenant cluster, that breaks isolation between tenants' agent workloads, with high confidentiality impact (scope changed). This is related to CVE-2026-100974, the shared secret itself, but here no special privileges are needed.
Who can reach it
Low-privileged process inside one OpenShell Kubernetes sandbox, with network reach to other sandboxes on the adjacent network (AV:A). No user interaction needed.
What to do
Update NVIDIA OpenShell to the fixed release in NVIDIA product-security bulletin 5929 and redeploy the sandboxes. The record does not give a fixed version. Rotate the handshake secret afterwards, since any sandbox could have read it. Until then, use Kubernetes NetworkPolicy to limit traffic between sandbox pods.
References
Related entries
- ONNX: Directory traversal via `external_data` field in the tensor protoCVE-2022-25882 · ONNXHigh
- Dagster (webserver): Directory traversalCVE-2023-51232 · Dagster (webserver)High
- Ray (log API): LFI — read any file on the head node, unauthenticatedCVE-2023-6021 · Ray (log API)High
- LiteLLM: Unauthenticated DoS via `ast.literal_eval` on user inputCVE-2024-10188 · LiteLLMHigh
- MLflow (`_create_model_version`): Path traversal in model-version creationCVE-2024-1558 · MLflow (`_create_model_version`)High
- Gradio (`/component_server`): Arbitrary method invocation on componentsCVE-2024-1561 · Gradio (`/component_server`)High
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.