PT-2026-95545
CVE-2026-33625
·
Published
2026-09-18
·
Updated
2026-10-01
Description by
CVSS 3.1
8.8
HighMetricsAV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H
Name of the Vulnerable Software and Affected Versions
LMDeploy versions 0.12.1 through 0.12.2
Description
Code injection is possible when loading a malicious HuggingFace model. The issue occurs because the
quant dtype value from the model's quantization config is passed to the eval() function in lmdeploy/pytorch/config.py without validation. An attacker can exploit this by publishing a model with a crafted quantization config.quant dtype variable, leading to arbitrary Python code execution on the machine loading the model. This represents a supply-chain attack vector affecting users who load untrusted models in local, cloud, or production environments.Recommendations
Update to version 0.12.3.
Found an issue in the description? Have something to add? Feel free to write us 👾