LMDeploy vulnerable to arbitrary code execution via eval() of untrusted quant_dtype in model config loading

Date Sep 18, 2026
Type Vendor advisory
Signal High-risk advisory
Vendor / Product PyPI ยท lmdeploy
CVE CVE-2026-33625
High-risk advisory CVE-2026-33625

Summary

LMDeploy vulnerable to arbitrary code execution via eval() of untrusted quant_dtype in model config loading

Why it matters

GitHub has published a reviewed security advisory for lmdeploy. Prioritize it with other application dependency updates, especially because EPSS places it in the 16% percentile.

Operator check

Check whether lmdeploy is present in application dependency manifests, lockfiles, or build images. Review the GitHub advisory and upgrade to a patched version where available. EPSS probability: 0.2%; percentile: 16%.

Sources

PatchBrief uses public sources. It does not scan environments, verify exposure, or replace vendor guidance.