CVE-2026-34755
Medium · 6.5CVSS
6.5
EPSS
0.8%
Published
Apr 6, 2026
CISA due date
—
The record
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 packages
| Ecosystem | Package | Introduced | Fixed |
|---|
Related vulnerabilities
- CVE-2026-114731 Roundcube Webmail before 1.6.20 and 1.7.x before 1.7.5 is affected by an uncontrolled… CWE-770
- CVE-2026-96377 Allocation of Resources Without Limits or Throttling vulnerability in Drupal Combined… CWE-770
- CVE-2026-96365 Allocation of Resources Without Limits or Throttling vulnerability in Drupal Webform… CWE-770
- CVE-2026-114723 Roundcube Webmail before 1.6.20 and 1.7.x before 1.7.5 allows the ImageMagick resource… CWE-770
- CVE-2026-113811 Algorithm::UrataniTakeda versions from 0.1.5 before 0.1.7 for Perl allow memory… CWE-770
- CVE-2026-113795 pypdf is a free and open-source pure-python PDF library. CWE-770
- CVE-2026-112400 Pillow is a Python imaging library. CWE-770
- CVE-2026-111064 AIOHTTP is an asynchronous HTTP client/server framework for asyncio and Python. CWE-770