# vLLM uncontrolled resource consumption in multimodal processors

Source: https://junglewise.ai/threats/duplicate-advisory-vllm-0-7-3-before-0-31-0-dos-via-unvalidated-mm-32363cc5  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-10-21. Updated: 2026-10-23  
License: CC BY 4.0 with a link back

Severity: high · CVSS 7.5 · Published 2026-10-21

vLLM is a large language model inference library used to serve AI models over APIs. The vulnerability allows remote API clients to specify parameters that control image processing (size, patch count, crop count, frame count), enabling attackers to force excessive memory allocation and crash the shared service, denying it to all users.

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 This vulnerability exploited in the wild?** No. Junglewise Threat Intelligence has no report of This vulnerability being exploited in the wild.

**How severe is This vulnerability?** High severity, with a CVSS score of 7.5, published 21 October 2026.

**Which products and versions are affected?** vllm (PyPI) >= 0.7.3, < 0.31.0, ECOSYSTEM:introduced=0.7.3; vLLM Project vllm 0.7.3 to before 0.31.0.

**Is there a fix for This vulnerability?** Yes: fixed in vLLM 0.31.0. Until then: Reject non-empty per-request mm_processor_kwargs at the API request schema by default. Operators serving only trusted clients can explicitly opt in to the existing parameter-override behavior; server-level --mm-processor-kwargs remains available for legitimate cases.</anigation>
<parameter name="timeline">[{"kind":"disclosed","at":"2026-10-06"},{"kind":"advisory","at":"2026-10-21"},{"kind":"patched","at":"2026-10-21","note":"Fix merged in PR #58830"}] Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.

**How is This vulnerability attacked?** It is attacked over the network, in a software library.

**Am I affected?** 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

- Fixed in vLLM 0.31.0

Reject non-empty per-request mm_processor_kwargs at the API request schema by default. Operators serving only trusted clients can explicitly opt in to the existing parameter-override behavior; server-level --mm-processor-kwargs remains available for legitimate cases.</anigation>
<parameter name="timeline">[{"kind":"disclosed","at":"2026-10-06"},{"kind":"advisory","at":"2026-10-21"},{"kind":"patched","at":"2026-10-21","note":"Fix merged in PR #58830"}]

## Technical details

Uncontrolled resource consumption (CWE-400) in multimodal image processors (Isaac, InternVL, H2OVL, others) where per-request mm_processor_kwargs are forwarded without validation to Hugging Face processor constructors. Remote, unauthenticated attackers over the network can set resource-sensitive parameters to exhaust GPU/CPU memory and terminate the server process. The vulnerability was introduced in v0.7.3 and fixed in v0.31.0.

## Affected products

- vLLM project vLLM 0.7.3 to before 0.31.0

## References

- [https://github.com/vllm-project/vllm/security/advisories/GHSA-gx7p-2j49-hfq4](https://github.com/vllm-project/vllm/security/advisories/GHSA-gx7p-2j49-hfq4)

## 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)