CVE-2026-37237

vllm: vLLM: Denial of Service via memory exhaustion from oversized media files

Published Aug 28, 2026 CVSS 3.1 · 7.5 High EPSS < 1% KEV · Not listed
Description
vLLM up to and including 0.17.0 allows remote attackers to cause a Denial of Service via memory exhaustion. The AsyncMediaIO.fetch_audio and AsyncMediaIO.fetch_image functions in multimodal/inputs.py fetch user-supplied media URLs using aiohttp and call r.read() without enforcing a maximum response size, allowing an attacker to exhaust server memory by providing a URL to an arbitrarily large file.

Analysis

AI-generated

Technical impact

The vulnerability resides in vLLM's media input handling functions. These functions fetch media from URLs provided by users and read the entire response into memory without imposing a maximum size. An attacker can supply a link to an enormous file, forcing the process to allocate excessive memory and eventually crash or become unresponsive. The weakness is a resource‑exhaustion flaw, which can lead to a denial‑of‑service condition for all users.

Affected Systems

vLLM library version 0.17.0 and earlier are vulnerable. No specific vendor name is available through a CNA. The flaw is present in the multimodal/inputs.py module that handles audio and image retrieval.

Risk and Exploitability

The vulnerability can be exploited remotely by any user who can submit URLs to the media fetching functions. While no public exploit or proof of concept is cited, the lack of a size restriction makes the attack straightforward once the vulnerable endpoint is reachable. The CVSS score of 7.5 indicates moderate‑to‑high severity. The EPSS score is less than 1%, and the vulnerability is not listed in the CISA KEV catalog, indicating no current widespread exploitation. Nevertheless, the potential to consume server memory and cause a service halt warrants prompt mitigation.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Upgrade vLLM to the latest version that imposes a maximum download size
  • Configure a reverse proxy or firewall to restrict the size of incoming media requests
  • Apply container or host‑level resource limits (memory, CPU) to protect the service from exhaustion

Generated by OpenCVE AI on September 1, 2026 at 17:44 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories available
History
Type Values Removed Values Added
First Time appeared Vllm
Vllm vllm
CPEs cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*
Vendors & Products Vllm
Vllm vllm
Type Values Removed Values Added
Title vllm: vLLM: Denial of Service via memory exhaustion from oversized media files
Weaknesses CWE-770
References
Metrics threat_severity

None

threat_severity

Important

Type Values Removed Values Added
Title Denial of Service via Unbounded Media Download in vLLM 0.17.0
Type Values Removed Values Added
Metrics cvssV3_1

{'score': 7.5, 'vector': 'CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H'}

ssvc

{'options': {'Automatable': 'yes', 'Exploitation': 'none', 'Technical Impact': 'partial'}, 'version': '2.0.3'}

Type Values Removed Values Added
Title Denial of Service via Unbounded Media Download in vLLM 0.17.0
Weaknesses CWE-400
Type Values Removed Values Added
First Time appeared Vllm-project
Vllm-project vllm
Vendors & Products Vllm-project
Vllm-project vllm
Type Values Removed Values Added
Description vLLM up to and including 0.17.0 allows remote attackers to cause a Denial of Service via memory exhaustion. The AsyncMediaIO.fetch_audio and AsyncMediaIO.fetch_image functions in multimodal/inputs.py fetch user-supplied media URLs using aiohttp and call r.read() without enforcing a maximum response size, allowing an attacker to exhaust server memory by providing a URL to an arbitrarily large file.
References
Weaknesses
  • CWE-400

    Uncontrolled Resource Consumption

  • CWE-770

    Allocation of Resources Without Limits or Throttling

Sources

Status
PUBLISHED
Assigner
mitre
Published
Updated
Aug 28, 2026, 7:30 PM
Reserved
Apr 6, 2026, 12:00 AM
Status
Analyzed
Published
Aug 28, 2026, 4:17 PM
Modified
Sep 29, 2026, 12:52 PM
Severity
Important
Public date
Aug 28, 2026, 12:00 AM