Is CVE-2026-105758 exploited in the wild?

No. Junglewise Threat Intelligence has no report of CVE-2026-105758 being exploited in the wild.

How severe is CVE-2026-105758?

Medium severity, with a CVSS score of 5.3, published 5 October 2026.

Which products and versions are affected?

vllm (PyPI) >= 0.24.0, < 0.30.0, ECOSYSTEM:introduced=0.24.0,fixed=0.30.0; Vllm 0.24.0 to 0.30.0.

Is there a fix for CVE-2026-105758?

Yes. vllm (PyPI): upgrade to 0.30.0 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.

How is CVE-2026-105758 attacked?

It is attacked over the network, in a software library.

Am I affected by CVE-2026-105758?

Check whether you run vllm (PyPI) or 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.

CVE-2026-105758: vLLM Qwen video backend memory exhaustion via unbounded sampling parameters

Q: Is there a fix for CVE-2026-105758?

Yes. vllm (PyPI): upgrade to 0.30.0 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.

Q: How is CVE-2026-105758 attacked?

It is attacked over the network, in a software library.

Q: Am I affected by CVE-2026-105758?

Check whether you run vllm (PyPI) or 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.

CVE-2026-105758 · Severity: medium · CVSS 5.3 · Published 2026-10-05

vLLM is an inference and serving engine for large language models that processes video input. An unauthenticated attacker can submit unbounded values for video frame sampling parameters in API requests, causing the service to consume excessive memory and crash before other requests can be processed. This could disrupt service availability for legitimate users.

Technologies: vllm (PyPI), Vllm. Vendors: PyPI, Vllm.

Quick answers

Is CVE-2026-105758 exploited in the wild?: No. Junglewise Threat Intelligence has no report of CVE-2026-105758 being exploited in the wild.
How severe is CVE-2026-105758?: Medium severity, with a CVSS score of 5.3, published 5 October 2026.
Which products and versions are affected?: vllm (PyPI) >= 0.24.0, < 0.30.0, ECOSYSTEM:introduced=0.24.0,fixed=0.30.0; Vllm 0.24.0 to 0.30.0.
Is there a fix for CVE-2026-105758?: Yes. vllm (PyPI): upgrade to 0.30.0 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.
How is CVE-2026-105758 attacked?: It is attacked over the network, in a software library.
Am I affected by CVE-2026-105758?: Check whether you run vllm (PyPI) or 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 0.30.0 or later
```
pip install "vllm>=0.30.0"
```

Technical details

The Qwen2VLVideoBackend and Qwen3VLVideoBackend classes in vLLM versions 0.24.0–0.30.0 accept and process request-level media_io_kwargs parameters for video.max_frames and video.fps without enforcing server-side ceilings. An unauthenticated network attacker can send crafted requests to the /tokenize endpoint with large values for these parameters, triggering excessive frame decoding that exhausts frontend memory and terminates the API process; the Rust frontend is not vulnerable because it rejects the media_io_kwargs field outright.

Affected products

vLLM vLLM 0.24.0 to 0.30.0

Identifiers

CVE-2026-105758
GHSA-x6mc-67gf-chw4
PYSEC-2026-4310

Timeline

2026-10-05: disclosed
2026-09-14: patched: Fix merged in commit ea723c8

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.