Code Injection Vulnerability in LMDeploy by InternLM
CVE-2026-33625

8.8HIGH

Key Information:

Vendor

Internlm

Status
Lmdeploy
Vendor
CVE Published:
18 September 2026

What is CVE-2026-33625?

LMDeploy, a toolkit designed for deploying large language models, is susceptible to a code injection vulnerability found in lmdeploy/pytorch/config.py. This flaw allows malicious actors to execute arbitrary Python code by manipulating the quantization_config.quant_dtype parameter of a malicious HuggingFace model. When the model is subsequently loaded using LMDeploy, the crafted quant_dtype bypasses necessary validation and is passed directly to eval(f'torch.{quant_dtype}'), exposing users to potential security threats. A fix for this issue has been implemented in version 0.12.3.

Affected Version(s)

lmdeploy >= 0.12.1, < 0.12.3

References

https://github.com/InternLM/lmdeploy/security/advisories/...
x_refsource_CONFIRM
https://github.com/InternLM/lmdeploy/releases/tag/v0.12.3
x_refsource_MISC

CVSS V3.1

Score:
8.8
Severity:
HIGH
Confidentiality:
High
Integrity:
High
Availability:
High
Attack Vector:
Network
Attack Complexity:
Low
Privileges Required:
None
User Interaction:
Required
Scope:
Unchanged

Timeline

  • Vulnerability published

  • Vulnerability Reserved

.