# Remediation Handoff: CVE-2026-105758

## Summary

NVD: vLLM is an inference and serving engine for large language models. NVD: From 0.24.0 until 0.30.0, the Qwen2VLVideoBackend and Qwen3VLVideoBackend classes accept request-level values for the media_io_kwargs.video.max_frames and media_io_kwargs.video.fps fields without enforcing server-side ceilings. NVD: An unauthenticated caller can submit these values to the /tokenize endpoint, causing the sampler to decode every frame selected from attacker-controlled video input, consume disproportionate frontend memory, and potentially terminate the API process before...

## Why it matters

CVSS severity is MEDIUM with score 5.3. EPSS percentile is 0.2128. KEV status is not listed in KEV. Use this as a defensive priority handoff, not as an execution instruction.

## Affected context

* Vendor: vllm
* Product: vllm
* Component: unknown
* Version / CPE / PURL: unknown

## Remediation context

* Recommended route: vendor_patch_or_mitigation
* Vendor/advisory reference: official_reference_present
* Patch status: confirmed
* Fixed version: 0.30.0
* References: https://github.com/vllm-project/vllm/commit/ea723c81c3ea26425cb69503a5d5e90822a04a45, https://github.com/vllm-project/vllm/pull/56729, https://github.com/vllm-project/vllm/releases/tag/v0.30.0, https://github.com/vllm-project/vllm/security/advisories/GHSA-x6mc-67gf-chw4

## Human checklist

* Confirm whether the listed product or package is present.
* Confirm affected version.
* Review vendor advisory or official source.
* Confirm patch, fixed version, mitigation, workaround, or monitor-only decision.
* Confirm exposure.
* Document remediation status.

## Safe AI / Codex handoff

* Goal: Prepare a defensive remediation plan for CVE-2026-105758 using only provided public-safe source context and any separate repo context supplied by the user.
* Allowed actions: summarize vendor guidance, prepare a defensive remediation plan, identify affected dependencies only when repo context is separately provided by the user, suggest tests and validation steps, document human verification questions
* Disallowed actions: generate offensive code, provide payloads, scan external targets, change production systems, merge or deploy changes, create GitHub issues or pull requests without explicit separate approval
* Acceptance criteria: Affected product or dependency presence is confirmed by a human., Affected version is confirmed or marked not applicable., Official advisory or source reference is reviewed., Patch, fixed version, mitigation, workaround, or monitor-only decision is documented., Validation steps are proposed without external scanning or production mutation.
* Human approval required: true

## Safety note

This handoff is for defensive triage and remediation planning only. It excludes offensive procedure, payload material, external target scan guidance, and auto-remediation.