Description
vLLM is an inference and serving engine for large language models (LLMs). From 0.7.0 to before 0.19.0, the VideoMediaIO.load_base64() method at vllm/multimodal/media/video.py splits video/jpeg data URLs by comma to extract individual JPEG frames, but does not enforce a frame count limit. The num_frames parameter (default: 32), which is enforced by the load_bytes() code path, is completely bypassed in the video/jpeg base64 path. An attacker can send a single API request containing thousands of comma-separated base64-encoded JPEG frames, causing the server to decode all frames into memory and crash with OOM. This vulnerability is fixed in 0.19.0.
Affected products
vllm: vllm
References10
- github.com/vllm-project/vllm/security/advisories/GHSA-pq5c-rjhq… PatchVendor Advisory
- access.redhat.com/errata/RHSA-2026:36005
- access.redhat.com/errata/RHSA-2026:36006
- access.redhat.com/errata/RHSA-2026:57380
- access.redhat.com/errata/RHSA-2026:57387
- access.redhat.com/errata/RHSA-2026:57389
- access.redhat.com/errata/RHSA-2026:57390
- access.redhat.com/errata/RHSA-2026:59144
- access.redhat.com/errata/RHSA-2026:59151
- access.redhat.com/security/cve/CVE-2026-34755
Metrics
- CVSS 3.1
-
6.5
MEDIUM
Score source: CNA
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H - EPSS — exploit probability
- 0% percentile: 42 · as of 12 November 2026
- CWE
- CWE-770
Could this be exploited in your environment?
A patch closes the known hole. A penetration test shows what an attacker can still reach — with proof.