Malicious attack method on hosted ML models now targets PyPI

ReversingLabs researchers discovered a short-lived PyPI supply-chain campaign delivering malicious Python packages that load Pickle-formatted PyTorch models which execute an infostealer at install time; the payload gathers system and .gitconfig data (likely targeting developers in China), was downloaded ~1,600 times across three packages, and highlights gaps in traditional security tooling for detecting executable behavior in ML model files.