CVE-2026-34755

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.

Affected Software

VendorProductVersion RangeStatus
vllm-projectvllm>= 0.7.0, < 0.19.0affected

Weaknesses

  • CWE-770: CWE-770: Allocation of Resources Without Limits or Throttling

ADP Enrichment

CISA ADP Vulnrichment

  • SSVC:
  • Exploitation: none
    • Automatable: no
    • Technical Impact: partial

vLLM: vLLM: Denial of Service due to excessive video frame processing

Additional References

References