CVE-2026-34755: vLLM: Denial of Service via Unbounded Frame Count in video/jpeg Base64 Processing

April 3, 2026 (updated April 6, 2026)

The VideoMediaIO.load_base64() method at vllm/multimodal/media/video.py:51-62 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 at line 47-48, 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.

References

github.com/advisories/GHSA-pq5c-rjhq-qp7p
github.com/vllm-project/vllm
github.com/vllm-project/vllm/commit/58ee61422169ce17e08248f8efa1e9df434fe395
github.com/vllm-project/vllm/pull/38636
github.com/vllm-project/vllm/security/advisories/GHSA-pq5c-rjhq-qp7p
nvd.nist.gov/vuln/detail/CVE-2026-34755

Code Behaviors & Features

Detect and mitigate CVE-2026-34755 with GitLab Dependency Scanning

Secure your software supply chain by verifying that all open source dependencies used in your projects contain no disclosed vulnerabilities. Learn more about Dependency Scanning →