Critical vulnerabilities in Hugging Face’s Diffusers library allow attackers to execute arbitrary code by bypassing safety mechanisms designed to protect against malicious AI model files. The flaws enable remote code execution (RCE) when users load compromised models from repositories, turning trust