# CVE-2026-33625: LMDeploy arbitrary code execution via eval() of untrusted quant_dtype

Source: https://junglewise.ai/threats/cve-2026-33625-lmdeploy-arbitrary-code-execution-via-eval-of-untrusted-quant  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-09-18. Updated: 2026-09-25  
License: CC BY 4.0 with a link back

CVE-2026-33625 · Severity: high · CVSS 8.8 · Published 2026-09-18

LMDeploy is a toolkit for deploying large language models. An attacker can trigger remote code execution by publishing a malicious model on HuggingFace Hub that contains a crafted quantization config. When a user loads the model with LMDeploy, arbitrary Python code embedded in the config is executed with the user's privileges, enabling full system compromise.

Technologies: [lmdeploy (PyPI)](https://junglewise.ai/threats/technologies/lmdeploy). Vendors: [PyPI](https://junglewise.ai/threats/vendors/pypi).

## Quick answers

**Is CVE-2026-33625 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-33625 being exploited in the wild.

**How severe is CVE-2026-33625?** High severity, with a CVSS score of 8.8, published 18 September 2026.

**Which products and versions are affected?** lmdeploy (PyPI) >= 0.12.1, < 0.12.3, ECOSYSTEM:introduced=0.12.1,fixed=0.12.3.

**Is there a fix for CVE-2026-33625?** Yes. lmdeploy (PyPI): upgrade to 0.12.3 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.

**How is CVE-2026-33625 attacked?** It is attacked over the network.

**Am I affected by CVE-2026-33625?** Check whether you run lmdeploy (PyPI) at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

## How to fix

- lmdeploy (PyPI): upgrade to 0.12.3 or later

  ```sh
  pip install "lmdeploy>=0.12.3"
  ```


## Technical details

The vulnerability is a code injection flaw in lmdeploy/pytorch/config.py line 620, where the quant_dtype parameter from a model's quantization_config is passed directly to eval(f'torch.{quant_dtype}') without validation. The quant_dtype value originates from a HuggingFace model's configuration file and is not sanitized, even though a similar torch_dtype parameter has a whitelist check. An attacker can craft a model with a quant_dtype value containing arbitrary Python code (e.g., "float16, __import__('os').system('id')"). The attack requires only that a user loads the malicious model via LMDeploy's standard APIs; no user interaction beyond that is needed, making this a supply-chain attack vector.

## Affected products

- InternLM LMDeploy all versions up to and including latest

## Identifiers

- CVE-2026-33625
- GHSA-3hmm-rh5q-gwwr
- PYSEC-2026-4078

## Timeline

- 2026-09-18: disclosed

## Related threats

- [LMDeploy remote code execution via pickle deserialization in DistServe](https://junglewise.ai/threats/cve-2025-66455-lmdeploy-remote-code-execution-via-pickle-deserialization-in)
- [LMDeploy SSRF bypass in URL validation](https://junglewise.ai/threats/lmdeploy-ssrf-bypass-in-url-validation-c4a40a8d)
- [LMdeploy remote code execution via pickle deserialization in zmq_rpc](https://junglewise.ai/threats/cve-2025-59953-lmdeploy-remote-code-execution-via-pickle-deserialization-in-zmq)
- [PYSEC-2026-1579 - InternLM LMDeploy code injection vulnerability](https://junglewise.ai/threats/cve-2025-3163-internlm-lmdeploy-code-injection-vulnerability)
- [PYSEC-2026-1577 - LMDeploy Improper Input Validation Vulnerability](https://junglewise.ai/threats/cve-2025-3162-lmdeploy-improper-input-validation-vulnerability)