Code Injection in PaddlePaddle PaddleX Preprocessor

Executive Summary

CVE-2026-108961 is an arbitrary code execution vulnerability in PaddlePaddle PaddleX up to version 3.7.2. It occurs in the NormalizeImage preprocessing step for text detection, where an attacker-controlled 'scale' string in the model directory's 'inference.yml' file is evaluated using Python's eval() function without proper validation. This allows malicious code execution during model loading.

Useful 0 Not Useful 0

Detection Guidance

Check for suspicious files named inference.yml in model directories, especially those containing malicious scale values like eval() calls or unexpected file writes. Inspect logs for unauthorized file creation or unusual model loading behavior.

Useful 0 Not Useful 0

Impact Analysis

An attacker can execute arbitrary code on your system by tricking you into loading a malicious model directory. This could lead to data theft, system compromise, or further network infiltration. The attack requires only a single modified line in 'inference.yml' and leaves no error trace, making detection difficult.

Useful 0 Not Useful 0

Compliance Impact

This vulnerability could lead to unauthorized data access or modification, violating GDPR's integrity and confidentiality principles or HIPAA's security requirements for protected health information. Organizations using affected PaddleX versions may face compliance violations if exploited.

Useful 0 Not Useful 0

Mitigation Strategies

Apply the official patch from pull request #5212 to replace eval() with a restricted numeric parser. Avoid using untrusted model directories and validate all third-party models before deployment.

Useful 0 Not Useful 0