GPU VulnDB

Database/AI/ML frameworks & serving

vLLM: no per-request media budget lets one request with many image URLs exhaust the frontend

CVSS 8.7CVE-2026-130686AI/ML frameworks & servingcurated

Impact

vLLM sets no limit on media across a whole request. It downloads and decodes every remote image URL before it checks prompt length. A small request to /v1/messages or /v1/responses that lists many image URLs can use up the frontend's memory, CPU and outbound connections, which denies service to everyone using that server. The record describes resource exhaustion only, with no data exposure.

Who can reach it

Anyone who can send requests to a vLLM API endpoint serving a multimodal model. No authentication is needed unless the operator puts it in front.

What to do

Upgrade vLLM to 0.33.0 or later and restart the serving process. Until then, cap request body size and the number of image parts at a gateway, or block remote image URL fetching if your clients do not need it.

References

Related entries

All AI/ML frameworks & serving entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.