GHSA-3mwp-wvh9-7528: vLLM: Unauthenticated OOM Denial of Service via Unbounded `n` Parameter in OpenAI API Server
- Identifiers
- CVE-2026-34756GHSA-3mwp-wvh9-7528
- Published
- Record updated
- Affected
- vllm >= 0.1.0, < 0.19.0
- Fixed in
- 0.19.0
- released April 3, 2026, on the day of disclosure
- Known exploitation
- Not listed in the CISA Known Exploited Vulnerabilities catalog at the last check.
- EPSS
- 0.4%
Summary
The vLLM OpenAI-compatible API server has a denial-of-service flaw in the `n` parameter of `ChatCompletionRequest` and `CompletionRequest`. These Pydantic models set no upper bound on `n`, and `_verify_args` in `vllm/sampling_params.py` checks only the lower bound, so an unauthenticated attacker can send one request with a very large `n`. The engine in `vllm/v1/engine/async_llm.py` then fans the request out into millions of copies, blocking the asyncio event loop and driving memory use up until the OS OOM-killer terminates the process.
Mitigation
The source does not state a fix yet. Check the original advisory for updates.
Affected packages in the Exposure Registry
Matched by package name and ecosystem. Each entry shows whether the package delegates to a language model and how many tracked packages depend on it.
- vllmPyPILLM dependency since 2023-08-25 · 15 tracked dependents
Topics
Related items
- LowToken fraud is here and it’s BAD (YouTube)Similar attack · Simon Willison's Weblog
- HighCVE-2026-133161: Pydantic AI Harness ToolOutputLimits ANSI stripping slowed by quadratic regexSimilar attack · NVD/CVE Database
- HighCVE-2026-130686: vLLM resource exhaustion through many remote image URLs in one requestSimilar attack · NVD/CVE Database
- HighCVE-2026-130685: vLLM denial of service through out-of-range prompt token ids in /v1/completionsSimilar attack · NVD/CVE Database
- LowRented Oscillations: Multi-Tenant Data Center Workload Synchronization Attacks and MitigationSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)