CNNVD-202512-4101 Information

CNNVD ID

CNNVD-202512-4101

CVE-2025-14927

  • CNNVD Published: 2025-12-23

Description (Chinese)

Hugging Face Transformers是Hugging Face开源的一个用于定义最先进机器学习模型的框架,涵盖文本、视觉、音频和多模态模型,可用于推理和训练。 Hugging Face Transformers存在代码注入漏洞,该漏洞源于SEW-D中convert_config函数在使用用户提供的字符串执行Python代码前缺乏验证,可能导致代码注入和远程代码执行。

Description (English)

Hugging Face Transformers is a framework for defining the state-of-the-art machine learning model that covers text, visual, audio and multi-modular models that can be used for reasoning and training. There is a code-injecting loophole in Hugging Face Transformers, which stems from the lack of authentication in the SEW-D convert config function before the Python code is executed using a string provided by the user, which may lead to code-injection and remote code execution.

Hazard Level

Medium

Vulnerability Type

代码注入

Affected Vendor

Hugging Face

Published

2025-12-23

Last Modified

2026-02-24

References

https://www.zerodayinitiative.com/advisories/ZDI-25-1148/

Share on: