vLLM: Memory exhaustion via media_io_kwargs num_frames override
Summary
vLLM versions 0.18.0 up to 0.31.0 allow API clients to bypass the frame-count limit for inline video/jpeg input by setting num_frames in media_io_kwargs, for example to -1. The API frontend then decodes and stacks every frame of a large inline video in memory, which can exhaust memory and crash or OOM-kill the serving process. The attack requires network access to the API and, in the worst case, no authentication. The impact is limited to denial of service.
Affected
- vLLM 0.18.0 up to but not including 0.31.0
- Downstream builds such as Red Hat AI Inference Server and Red Hat OpenShift AI are listed as affected in a third-party record; check the vendor status for your build
Fixed in
- vLLM 0.31.0 (according to the CVE record and GHSA-4m6g-h9mv-wqjc; confirm the exact fixed version in the advisory)
- The fix rejects non-empty per-request media_io_kwargs by default (PR #58830)
Workaround
- Unknown. No vendor-documented workaround for this CVE.
- Until patched: do not expose the API to untrusted clients, put it behind authentication and a gateway with request-size limits, and make sure --trust-request-mm-kwargs is not enabled.
Exploitation
- Not listed as known exploited (no CISA KEV entry).
- No public proof of concept found in the sources checked.
- CISA SSVC: exploitation none, automatable yes, technical impact partial.
Our recommendation
- Inventory vLLM instances and versions, including container images and Red Hat AI/OpenShift AI deployments.
- Upgrade to vLLM 0.31.0 or later.
- Remove direct internet exposure of the API and enforce authentication.
- Verify that --trust-request-mm-kwargs is disabled on exposed servers.
- Monitor for API process restarts and memory spikes.
Public exploits
PoC-in-GitHub, Metasploit and Exploit-DB list nothing for this CVE.
Research sources
- GitHub Security Advisory GHSA-4m6g-h9mv-wqjc github.com
- VulnCheck advisory: vLLM memory exhaustion via media_io_kwargs num_frames override www.vulncheck.com
- OSV record via Vulners vulners.com
- OpenCVE: CVE-2026-116179 app.opencve.io
- HOL Guard: CVE-2026-116179 hol.org
- OffSeq Radar: CVE-2026-116179 radar.offseq.com
- vLLM documentation: multimodal media video (v0.18.0) docs.vllm.ai
Researched and written up in English and German by an AI (Claude by Anthropic) based on NVD, CISA, EPSS and the sources listed. The vendor information is authoritative. Researched on 1 Nov 2026, 00:18.
Original description (NVD)
vLLM from 0.18.0 before 0.31.0 contains a resource exhaustion vulnerability that allows API clients to bypass the video/jpeg frame-count cap by overriding num_frames through media_io_kwargs. Attackers can set num_frames to -1 with a large inline video so VideoMediaIO.load_base64 decodes every frame, exhausting memory and crashing the API frontend.
Sources
The tracker combines official, freely usable data sources – updated continuously:
- NVD (NIST) – CVEs, rating (CVSS), affected products
- CISA KEV – vulnerabilities confirmed by CISA as actively exploited
- CISA Vulnrichment – CISA assessment (SSVC) in the official CVE record
- FIRST EPSS – probability of exploitation
- Microsoft MSRC – Patch Tuesday: vulnerabilities, severity, updates
- PoC-in-GitHub – public proof-of-concept repositories
- Metasploit (Rapid7) – modules of the Metasploit Framework
- Exploit-DB (OffSec) – archive of public exploits (links only)
- Patch days: official vendor pages (linked per vendor) (Vendor patch days)
This product uses data from the NVD API but is not endorsed or certified by the NVD. EPSS data © FIRST.org.
Affected? We can help.
We check whether your systems are affected, install updates and close the gap – continuously with the System Health Checker if you wish.