AIKIDO-2026-897456
vllm is vulnerable to Denial of Service (DoS)
75
High Risk
This Affects:
vllmTL;DR
vLLM's VideoMediaIO.load_base64() method processes video/jpeg data URLs by splitting the base64 payload on commas to extract individual JPEG frames. It does not enforce any limit on the number of frames before decoding them into memory. A remote attacker can send a single unauthenticated request to the OpenAI-compatible chat completions API containing thousands of comma-separated frames, exhausting server memory and crashing the engine. The fix enforces a frame count limit when loading video media.
Who does this affect?
You are affected if you are using a version that falls within the vulnerable range and you serve models that accept multimodal video inputs over the OpenAI-compatible API.
Background info
vllm is vulnerable to Denial of Service (DoS) in versions 0.7.0 - 0.18.1.
How to fix this
Upgrade the vllm library to the patch version.
Links
GitHub Release
Other
huntr.com/bounties/7bd92629-b396-4449-8f88-6c0092530eb4access.redhat.com/security/cve/CVE-2026-5497bugzilla.redhat.com/show_bug.cgi?id=2487813security.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-5497.json
Are You Affected?
Connect your repositories to instantly see whether vulnerable or malicious packages exist in your codebase.
Free. No credit card required.
I consent to receiving marketing communications based on Aikido’s Privacy Policy.
SOC 2Compliant
ISO 27001Compliant