Description
vLLM through 0.29.0 fetches and fully materializes remote or inline media before enforcing its documented media controls (the VLLM_MAX_AUDIO_CLIP_FILESIZE_MB compressed-audio size cap, default 25 MB, and the per-modality --limit-mm-per-prompt item limits). Across four ingress paths — the shared media-acquisition layer (HTTPConnection.get_bytes()/async_get_bytes()), the chat completions audio_url/base64 path, the batch speech runner, and the Rust frontend POST /tokenize route — the server reads the entire HTTP response body, base64-decodes the inline payload, or spawns one fetch/decode task per media part, and only then applies the limit (or, on some paths, never applies it). A remote attacker can therefore cause the API server or batch-runner process to allocate memory and consume outbound bandwidth proportional to an attacker-chosen body size or media item count before the request is rejected, resulting in pre-inference memory and bandwidth exhaustion (denial of service). The chat and batch surfaces require an API key when one is configured; the Rust frontend /tokenize route is unauthenticated by design. There is no code execution or data disclosure impact.
Published: 2026-09-26
Score: 7.1 High
EPSS: n/a
KEV: No
Impact: Resource Exhaustion (Denial of Service)
Action: Patch
AI Analysis

Impact

vLLM versions prior to 0.29.0 download and fully materialize remote or inline media before validating size limits, leading to an attacker‑controlled allocation of memory and outbound bandwidth. The vulnerability is a classic Resource Exhaustion weakness, defined by CWE‑400. It does not lead to code execution or data disclosure, but it can exhaust server resources and disrupt inference workloads.

Affected Systems

The affected product is vLLM from the vllm‑project. All releases before 0.29.0 are susceptible. No exact patch version is listed, but the default behavior is present until the 0.29.0 release or later versions.

Risk and Exploitability

With a CVSS score of 7.1 the vulnerability is considered medium–high severity. The EPSS score is not available, and it is not listed in the CISA KEV catalog, indicating no known active exploitation yet. The most likely attack vector is remote requests to the media ingestion endpoints, which can be sent without an API key on the Rust frontend /tokenize route. A malicious actor could upload or request large media files, causing memory and bandwidth exhaustion before the request is rejected. The risk is elevated if the system is publicly exposed or lacks strict media size limits.

Generated by OpenCVE AI on September 26, 2026 at 16:29 UTC.

Remediation

No vendor fix or workaround currently provided.

OpenCVE Recommended Actions

  • Upgrade to vLLM 0.29.0 or later, which applies media size limits before fetching.
  • If an immediate upgrade is not possible, configure stricter media limits such as reducing VLLM_MAX_AUDIO_CLIP_FILESIZE_MB and the per‑modality limit to very small values or disabling media ingestion altogether.
  • Ensure the Rust frontend /tokenize route is protected by authentication or otherwise restricted, since it is currently unauthenticated by design.
  • Monitor server memory and bandwidth usage for abnormal spikes that could indicate an attempted resource exhaustion attack.

Generated by OpenCVE AI on September 26, 2026 at 16:29 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories yet.

History

Sat, 26 Sep 2026 13:30:00 +0000

Type Values Removed Values Added
Description vLLM through 0.29.0 fetches and fully materializes remote or inline media before enforcing its documented media controls (the VLLM_MAX_AUDIO_CLIP_FILESIZE_MB compressed-audio size cap, default 25 MB, and the per-modality --limit-mm-per-prompt item limits). Across four ingress paths — the shared media-acquisition layer (HTTPConnection.get_bytes()/async_get_bytes()), the chat completions audio_url/base64 path, the batch speech runner, and the Rust frontend POST /tokenize route — the server reads the entire HTTP response body, base64-decodes the inline payload, or spawns one fetch/decode task per media part, and only then applies the limit (or, on some paths, never applies it). A remote attacker can therefore cause the API server or batch-runner process to allocate memory and consume outbound bandwidth proportional to an attacker-chosen body size or media item count before the request is rejected, resulting in pre-inference memory and bandwidth exhaustion (denial of service). The chat and batch surfaces require an API key when one is configured; the Rust frontend /tokenize route is unauthenticated by design. There is no code execution or data disclosure impact.
Title vLLM before 0.29.0 Resource Exhaustion via Unbounded Media Materialization
First Time appeared Vllm
Vllm vllm
Weaknesses CWE-400
CPEs cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*
Vendors & Products Vllm
Vllm vllm
References
Metrics cvssV3_1

{'score': 6.5, 'vector': 'CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H'}

cvssV4_0

{'score': 7.1, 'vector': 'CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N'}

Subscriptions

Vllm Vllm
cve-icon MITRE

Status: PUBLISHED

Assigner: VulnCheck

Published:

Updated: 2026-09-26T13:23:20.418Z

Reserved: 2026-09-26T02:33:07.898Z

Link: CVE-2026-100650

cve-icon Vulnrichment

No data.

cve-icon NVD

Status : Received

Published: 2026-09-26T14:16:47.523

Modified: 2026-09-26T14:16:47.523

Link: CVE-2026-100650

cve-icon Redhat

No data.

cve-icon OpenCVE Enrichment

Updated: 2026-09-26T16:30:07Z

Weaknesses
  • CWE-400

    Uncontrolled Resource Consumption