預處理工具供應鏈漏洞

Risk Description

Security vulnerabilities in data pre-processing tools (image libraries like Pillow/OpenCV, text cleaning libraries, audio processing libraries) can be exploited to inject malicious code or trigger buffer overflow attacks. Since pre-processing occurs before model inference, attackers can trigger tool vulnerabilities through crafted input data, gaining system control before the AI system processes any request.