7.5highHigh

CVE-2026-41523

Vllm Vllm

vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.0, an assert-based security check in vLLM's activation function loading allows any unauthenticated attacker to achieve arbitrary code execution on the server by publishing a malicious HuggingFace model, when vLLM runs in Python optimized mode (python -O or PYTHONOPTIMIZE=1). This vulnerability is fixed in 0.22.0.

Exploitation status

Published references include exploit or proof-of-concept material. There is no confirmation it is being used in real attacks.
A vendor advisory or patch reference has been published.

What the metrics mean

It is reachable over a network, without needing local access.
For an attacker who can reach it, exploitation depends on conditions outside the attacker's control.
To exploit it, no account is needed.
Beyond that, someone has to be persuaded to take an action first.

Scoring

CVSS: 7.5 (v3.1)
Vector: CVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:U/C:H/I:H/A:H
Weakness: CWE-94
Assigned by: security-advisories@github.com

Dates

Published: 2026-06-22
Last modified: 2026-09-04
Sources: NVD

Affected products

Vllm Vllm- 0.22.0

As listed in the NVD configuration data. Not a statement about your estate.

Exploitation status

What the metrics mean

Scoring

Dates

Affected products

References