Malicious AI Models on Hugging Face Exploit Novel Attack Technique

Reversing Labs found two malicious ML models on Hugging Face that used 7z-compressed PyTorch (Pickle) files to bypass the platform's Picklescan detection, demonstrating a novel technique for embedding malicious code in 'broken' Pickle streams; Hugging Face removed the models within 24 hours and updated Picklescan to detect threats in broken Pickle files.