CVE-2026-34755: vLLM is an inference and serving engine for large language models (LLMs). From 0.7.0 to before 0.19.…
- Occurred
- Apr 6, 2026
- Published
- Apr 6, 2026
- Source
- NVD CVE
Summary
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.
Tags
- CVE
- CVE-2026-34755
- cvss:MEDIUM
Classification
Rationale: vLLM is an LLM serving engine with a DoS vulnerability allowing frame count bypass causing OOM crashes.
Classifier: Qwen/Qwen3.5-35B-A3B-FP8