Researchers discovered two malicious ML models on Hugging Face exploiting “broken” pickle files to evade detection, bypassing ...