# CVE-2026-116179: vLLM resource exhaustion in VideoMediaIO frame-count validation

Source: https://junglewise.ai/threats/cve-2026-116179-vllm-from-0-18-0-before-0-31-0-contains-a-resource-exhaustion  
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-2026-116179 · Severity: high · CVSS 7.5 · Published 2026-10-25

vLLM is a framework for running large language models with multimodal support. An API client can bypass the video frame-count limit by sending a specially crafted request, causing the API server to decode and buffer an entire large video file in memory. This can exhaust server memory and crash the service, enabling a denial-of-service attack.

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

## Quick answers

**Is CVE-2026-116179 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-116179 being exploited in the wild.

**How severe is CVE-2026-116179?** High severity, with a CVSS score of 7.5, published 25 October 2026.

**Which products and versions are affected?** vLLM Project vllm >=0.18.0, <0.31.0.

**Is there a fix for CVE-2026-116179?** Yes: fixed in 0.31.0. Until then: Deployments that trust their API clients can enable --trust-request-mm-kwargs to opt in to per-request media_io_kwargs support; the default configuration now rejects non-empty per-request media_io_kwargs to prevent client overrides of video decoder settings. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.

**How is CVE-2026-116179 attacked?** It is attacked over the network, in a software library.

**Am I affected by CVE-2026-116179?** Check whether you run 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 0.31.0

Deployments that trust their API clients can enable --trust-request-mm-kwargs to opt in to per-request media_io_kwargs support; the default configuration now rejects non-empty per-request media_io_kwargs to prevent client overrides of video decoder settings.

## Technical details

The VideoMediaIO.load_base64 method enforces a frame-count cap for video/jpeg data, but this limit can be overridden via request-level media_io_kwargs by setting num_frames to -1, which triggers unlimited frame decoding. An unauthenticated attacker with network access to the chat endpoint can submit a large inline video payload that causes the API frontend to allocate gigabytes of memory before validation rejects the request, resulting in denial of service. A fix was released in version 0.31.0.

## Affected products

- vLLM Project vLLM >=0.18.0, <0.31.0

## Timeline

- 2026-10-09: disclosed
- 2026-10-25: advisory

## References

- [https://github.com/vllm-project/vllm/security/advisories/GHSA-4m6g-h9mv-wqjc](https://github.com/vllm-project/vllm/security/advisories/GHSA-4m6g-h9mv-wqjc)
- [https://www.vulncheck.com/advisories/vllm-0.18.0-before-0.31.0-memory-exhaustion-via-media-io-kwargs-num-frames-override](https://www.vulncheck.com/advisories/vllm-0.18.0-before-0.31.0-memory-exhaustion-via-media-io-kwargs-num-frames-override)

## 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 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 Phi-4 multimodal audio encoder GPU memory exhaustion](https://junglewise.ai/threats/cve-2026-115913-vllm-is-an-inference-and-serving-engine-for-large-language)
- [vLLM arbitrary code execution in shared-memory multimodal cache](https://junglewise.ai/threats/cve-2026-115912-vllm-is-an-inference-and-serving-engine-for-large-language)