CVE-2026-34755

Low

No strong exploitation signal.

CVSS base

6.5 MEDIUM

CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H

EPSS — probability of exploitation (30 days)

0.8%

56.2th percentile

CISA KEV

Not listed

Weakness / dates

CWE-770

Published 2026-04-06 · modified 2026-08-25

CVSS breakdown

CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H

Attack Vector	N	Network
Attack Complexity	L	Low
Privileges Required	L	Low
User Interaction	N	None
Scope	U	Unchanged
Confidentiality	N	None
Integrity	N	None
Availability	H	High

Timeline

2026-04-06 — Published (NVD)
2026-08-25 — Last modified (NVD)

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.

Affected

vllm

References

Official: NVD · CVE.org