UTC
High

Denial of service: a negative token id in `/v1/embeddings` or `/pooling` input kills the vLLM engine via a CUDA device-side assertion

CVE-2026-93592 · GHSA-25q3-v2hm-8vpf · published 2026-09-03 · updated 2026-09-30

Severity
High
CVSS score
7.5
Chance of exploitation (EPSS)
Not scored
Fixed in
0.28.0
  • Exploitation Public proof of concept
  • Automatable Yes
  • Technical impact Partial
  • CISA SSVC
What it is

Any deployment that serves a pooling or embedding model through the OpenAI-compatible API server. One unauthenticated HTTP request is a complete denial of service for that server process and every client using it. Restarting under a supervisor does not mitigate it, because the same request can be sent again immediately. Confidentiality and integrity are unaffected; the impact is availability only.

What to do

Upgrade vllm to 0.28.0 or later. If you start it from an MCP config with npx or uvx, pin that version so an old cached copy is not used.

Fixed in 0.28.0 by https://github.com/vllm-project/vllm/pull/51795 (commit b216db3ed0), which adds the missing lower-bound check to InputProcessor._validate_model_input so a negative id is rejected as HTTP 400 before the ids reach the device: The validation path is shared by generate, embedding and pooling requests, so the check covers all three.

If you cannot upgrade yet

For deployments that cannot upgrade to 0.28.0: Reject request bodies containing negative integers in input at a reverse proxy or API gateway in front of vLLM. Do not expose the token-id form of the input; accept only string input from untrusted callers. Restrict network access to /v1/embeddings and /pooling. Process supervision is not a workaround: the request can simply be repeated.

Affected
PackageRegistryAffectedFixed in
vllmPyPI< 0.28.00.28.0

The impact, patch and workaround text comes from the advisory, shortened (published by vLLM). Proof-of-concept and reproduction sections and code are left out; the original advisory has them.

Sources: GitHub Advisory Database, reviewed and malware advisories (CC BY 4.0, https://github.com/advisories); advisories published by Cursor (https://github.com/cursor/cursor/security/advisories); advisories published by OpenClaw (https://github.com/openclaw/openclaw/security/advisories); advisories published by vLLM (https://github.com/vllm-project/vllm/security/advisories); advisories published by Open WebUI (https://github.com/open-webui/open-webui/security/advisories); advisories published by n8n (https://github.com/n8n-io/n8n/security/advisories); advisories published by Flowise (https://github.com/FlowiseAI/Flowise/security/advisories); advisories published by Langflow (https://github.com/langflow-ai/langflow/security/advisories); advisories published by the MCP Python SDK project (https://github.com/modelcontextprotocol/python-sdk/security/advisories); advisories published by MLflow (https://github.com/mlflow/mlflow/security/advisories); CISA Known Exploited Vulnerabilities catalog (CC0). Data as of 2026-11-22. ShieldMap describes how each problem works and how to fix it; it does not publish exploit code. Report a correction to [email protected].