# CVE-2026-116173: vLLM Rust gRPC denial of service via output_candidates validation bypass

Source: https://junglewise.ai/threats/cve-2026-116173-vllm-from-0-22-0-before-0-24-0-contains-a-denial-of-service  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-10-25. Updated: 2026-10-25  
License: CC BY 4.0 with a link back

CVE-2026-116173 · Severity: medium · CVSS 5.9 · Published 2026-10-25

vLLM is an open-source large language model inference engine that serves AI models over a network interface. Attackers can crash the inference engine by sending specially crafted requests that bypass input validation in the Rust gRPC frontend, causing a denial of service that takes the model offline and blocks all inference requests.

Technologies: [vllm (PyPI)](https://junglewise.ai/threats/technologies/pypi-vllm), [vLLM Project vllm](https://junglewise.ai/threats/technologies/vllm-project-vllm). Vendors: [PyPI](https://junglewise.ai/threats/vendors/pypi), [vLLM Project](https://junglewise.ai/threats/vendors/vllm-project).

## Quick answers

**Is CVE-2026-116173 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-116173 being exploited in the wild.

**How severe is CVE-2026-116173?** Medium severity, with a CVSS score of 5.9, published 25 October 2026.

**Which products and versions are affected?** vllm (PyPI) ECOSYSTEM:introduced=0.22.0,fixed=0.24.0; vLLM Project vllm 0.22.0 to before 0.24.0.

**Is there a fix for CVE-2026-116173?** Yes. vllm (PyPI): upgrade to 0.24.0 or later. Until then: Restrict network access to the Rust gRPC endpoint to trusted clients only, or run vLLM behind an API gateway that validates output_candidates.token_ids length before forwarding requests. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.

**How is CVE-2026-116173 attacked?** It is attacked over the network.

**Am I affected by CVE-2026-116173?** Check whether you run vllm (PyPI) or vLLM Project vllm at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

## How to fix

- vllm (PyPI): upgrade to 0.24.0 or later

  ```sh
  pip install "vllm>=0.24.0"
  ```


Restrict network access to the Rust gRPC endpoint to trusted clients only, or run vLLM behind an API gateway that validates output_candidates.token_ids length before forwarding requests.

## Technical details

The Rust gRPC frontend accepts ResponseOptions.output_candidates.token_ids without enforcing the Python SamplingParams validator's 128-token limit (MAX_LOGPROB_TOKEN_IDS). Attackers send Generate or GenerateStream requests with output_logprobs enabled and over 128 token IDs, bypassing early validation and causing a fatal exception in the GPU logprob sampler when the oversized list reaches fixed-width tensor allocation. The vulnerability requires network access to the gRPC endpoint with no authentication required.

## Affected products

- vLLM Project vLLM 0.22.0 to before 0.24.0

## Identifiers

- CVE-2026-116173
- GHSA-6pgw-f3pv-4h8q
- PYSEC-2026-4350

## Timeline

- 2026-10-09: disclosed
- 2026-10-25: advisory
- 2026-10-25: patched: vLLM 0.24.0 released

## References

- [https://github.com/vllm-project/vllm/security/advisories/GHSA-6pgw-f3pv-4h8q](https://github.com/vllm-project/vllm/security/advisories/GHSA-6pgw-f3pv-4h8q)
- [https://www.vulncheck.com/advisories/vllm-0.22.0-before-0.24.0-denial-of-service-via-grpc-output-candidates-token-ids](https://www.vulncheck.com/advisories/vllm-0.22.0-before-0.24.0-denial-of-service-via-grpc-output-candidates-token-ids)

## Related threats

- [vLLM chat template resource exhaustion DoS](https://junglewise.ai/threats/cve-2026-115918-vllm-is-an-inference-and-serving-engine-for-large-language)
- [vLLM: Permissive default CORS origins allow cross-origin requests to the vLLM HTTP server](https://junglewise.ai/threats/cve-2026-115917-vllm-is-an-inference-and-serving-engine-for-large-language)
- [vLLM off-by-one error in tokenizer max_token_id validation](https://junglewise.ai/threats/cve-2026-115916-vllm-is-an-inference-and-serving-engine-for-large-language)
- [vLLM image format validation bypass in media processing](https://junglewise.ai/threats/cve-2026-115915-vllm-is-an-inference-and-serving-engine-for-large-language)
- [vLLM denial of service in DeepSeek-OCR image processing](https://junglewise.ai/threats/cve-2026-115914-vllm-is-an-inference-and-serving-engine-for-large-language)