CVE-2026-33625 LMDeploy vulnerable to arbitrary code execution via eval() of untrusted quant_dtype in model config loading
Attackers can execute arbitrary code by publishing malicious HuggingFace models with crafted quantization configs, affecting LMDeploy users.
- Exploited in the wild
- Not known
- Login needed
- Not needed
- Attackable
- Over the internet
- Victim action needed
- A user has to click or open something
- Fix
- Available
What is CVE-2026-33625?
CVE-2026-33625 is a high-severity code injection vulnerability in InternLM lmdeploy, published on 18 Sep 2026. An attacker can exploit it over the internet without logging in, and only if a user opens a file or follows a link.
Official description
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.
Weakness: CWE-400
Which versions are affected by CVE-2026-33625?
3 open source packages ship an affected version, including lmdeploy.
| Package | Affected | Fixed in |
|---|---|---|
lmdeploy PyPI | From 0.12.1 | 0.12.3 |
lmdeploy PyPI | From 0.12.1 | 0.12.3 |
lmdeploy PyPI | From 0.12.1 | 0.12.3 |
How do I fix CVE-2026-33625?
Upgrade lmdeploy to 0.12.3 or later.
Upgrade, and never evaluate templates, expressions or code built from user input. More about Code injection.
Is CVE-2026-33625 being exploited?
It is not confirmed as exploited in the wild, but public exploit code for CVE-2026-33625 exists.
Sources
Also known as GHSA-3hmm-rh5q-gwwr, PYSEC-2026-4078.