Database/AI/ML frameworks & serving
NVIDIA OpenShell: cleartext HTTP/2 (h2c) inside a CONNECT tunnel bypasses L7 REST policy
Impact
NVIDIA reports that the OpenShell proxy and the upstream server read HTTP differently when cleartext HTTP/2 (h2c) runs inside an HTTP CONNECT tunnel. A sandboxed workload can use that to send REST calls the policy should block. Confidentiality and integrity impact are high (scope changed), and attack complexity is rated high.
Who can reach it
Local attacker with low privileges inside an OpenShell sandbox. No user interaction needed.
What to do
Update NVIDIA OpenShell to the fixed release in NVIDIA product-security bulletin 5929. The record does not give a fixed version. Until then, consider turning off h2c on upstream services that sandboxes can reach.
References
Related entries
- NVIDIA OpenShell: CLI diagnostic command escaped for one shell layer but run through two, allowing command injectionCVE-2026-100980 · NVIDIA OpenShell for Linux (CLI diagnostic interface)High
- MLflow: schemeless prompt source lets a caller read arbitrary files readable by the tracking serverCVE-2026-116690 · MLflow model registry (_create_model_version prompt branch / get-artifact)High
- vLLM orchestrator gateway: bearer tokens and full chat payloads written to persistent logsCVE-2026-15574 · vllm-orchestrator-gateway (Red Hat OpenShift AI)High
- SGLang (`/server_info`): Endpoint returns API keys and SSL keyfile pathsCVE-2026-15977 · SGLang (`/server_info`)High
- SGLang (weight exfiltration): Two endpoints allow a remote attacker to pull model weights when no API key is setCVE-2026-15978 · SGLang (weight exfiltration)High
- NVIDIA Triton Inference Server: unauthenticated request triggers excessive iteration and hangs the serverCVE-2026-16497 · NVIDIA Triton Inference Server for LinuxHigh
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.