CVE-2026-34755
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.
AI Insight
LLM-synthesized narrative grounded in this CVE's description and references.
Affected packages
Versions sourced from the GitHub Security Advisory.
| Package | Affected versions | Patched versions |
|---|---|---|
vllmPyPI | >= 0.7.0, < 0.19.0 | 0.19.0 |
Affected products
6- osv-coords5 versionspkg:apk/chainguard/py3.10-vllm-cuda-12.4pkg:apk/chainguard/py3.12-vllm-cuda-12.4pkg:apk/chainguard/tritonserver-backend-vllm-cuda-13.0pkg:apk/chainguard/vllm-openai-cuda-12.9pkg:pypi/vllm
< 0.18.1-r2+ 4 more
- (no CPE)range: < 0.18.1-r2
- (no CPE)range: < 0.18.1-r2
- (no CPE)range: < 25.11-r7
- (no CPE)range: < 0.19.0-r0
- (no CPE)range: >= 0.7.0, < 0.19.0
Patches
Vulnerability mechanics
References
6- github.com/vllm-project/vllm/security/advisories/GHSA-pq5c-rjhq-qp7pnvdPatchVendor AdvisoryWEB
- github.com/advisories/GHSA-pq5c-rjhq-qp7pghsaADVISORY
- nvd.nist.gov/vuln/detail/CVE-2026-34755ghsaADVISORY
- github.com/pypa/advisory-database/tree/main/vulns/vllm/PYSEC-2026-144.yamlghsaWEB
- github.com/vllm-project/vllm/commit/58ee61422169ce17e08248f8efa1e9df434fe395ghsaWEB
- github.com/vllm-project/vllm/pull/38636ghsaWEB
News mentions
0No linked articles in our index yet.