← All advisories
CVE-2026-130685highDenial of service2026-11-18

vLLM before 0.33.0: one request can crash EngineCore or exhaust the frontend, and a malicious model config can swap in another local model (CVE-2026-130684/130685/130686)

Three vLLM CVEs fixed in 0.33.0 (released 2 November 2026) were published on 18 November. With --skip-tokenizer-init, one unauthenticated request carrying an out-of-range prompt token id crashes the shared EngineCore for every tenant. Any multimodal deployment can be hit by a single request listing many remote image URLs, all downloaded and decoded before length checks. And a served model whose config.json has speculators_config.verifier.name_or_path can redirect vLLM to another local checkpoint and leak its tokenizer via /detokenize.

01.

What it is

CVE-2026-130685 / GHSA-q9v9-qpjp-fc24 (vLLM 0.28.0 to before 0.33.0): the vocab-range check added for GHSA-25q3-v2hm-8vpf only runs when a tokenizer exists, so with --skip-tokenizer-init a large token id in /v1/completions (and /v1/embeddings, /pooling) reaches the embedding as an index and EngineCore shuts down. The OpenAI-compatible server has no auth by default.

CVE-2026-130686 / GHSA-5r3c-5x4v-8gjh (0.6.0 to before 0.33.0): there is no request-wide byte or memory budget for media, and the unspecified per-modality count defaults to 999. One compact request to /v1/messages or non-Harmony /v1/responses can tie up outbound connections, decoder CPU and API-frontend memory.

CVE-2026-130684 / GHSA-qrj2-j723-m6w8 (before 0.33.0): maybe_override_with_speculators runs for every non-cloud-storage model and takes speculators_config verifier.name_or_path from the artifact's config.json as both model and tokenizer with no containment check, even with trust_remote_code off. A malicious model publisher can point it at another local checkpoint directory.

02.

How to check

  1. Run pip show vllm (or check your image tag) and compare with 0.33.0.
  2. Extra exposure: --skip-tokenizer-init deployments (token-in/token-out, RL serving), multimodal models that accept remote image URLs, and servers that load models from third-party repositories.
03.

How to fix

  1. Upgrade vLLM to 0.33.0 or later.
  2. Put vLLM behind an authenticating gateway with request-size and rate limits; set --limit-mm-per-prompt to the image count you actually need.
  3. Only serve model repositories you trust, and check config.json for a speculators_config key before serving a third-party model.
pip show vllm | grep Version
# does a downloaded model try to redirect the verifier?
grep -n speculators_config /path/to/model/config.json
04.

Sources

Related on VibeVulns
More advisories for this tool
The weekly email

One email a week. Only what changed.

Every new security hole in AI coding tools, MCP servers, app builders and the stacks they generate (Next.js, Supabase, Node, PHP), each with how to check your app and the fix. No roundups of things you already read.

Free, one email a week. Read the latest issue first if you like. One-click unsubscribe in every email. We never sell or share addresses.