vulnerability

CVE-2026-33625

NVD HIGH 8.8 · published 2026-09-18 · 1 on the wire

LMDeploy is a toolkit for compressing, deploying, and serving large language models. Versions 012.1 through 0.12.2 contain a code injection vulnerability in `lmdeploy/pytorch/config.py` line 620 that allows an attacker to execute arbitrary Python code by publishing a malicious HuggingFace model with a crafted `quantization_config.quant_dtype` value. When a user loads the model with lmdeploy, the `quant_dtype` is passed to `eval(f'torch.{quant_dtype}')` without any validation. Version 0.12.3 contains a patch.

Exploited in the wild

Not on KEV

Not in CISA's Known Exploited Vulnerabilities catalogue as of the last daily pull. Absence is not proof of safety.

EPSS · 30-day forecast

—

FIRST has not scored this id yet.

CVSS · NVD

8.8

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

Internet exposure

—

No exposure census on this CVE's dispatches.

On the wire1

18 SepGHSA-3hmm-rh5q-gwwr: LMDeploy vulnerable to arbitrary code execution via eval() of untrusted quant_dtype in model config loadingVULN

References

Early access

Watch this one?

Early access opens alerts first — one email when a CVE you follow lands on KEV or an adversary you follow lands on the wire. Nothing else, ever.

bot-protected