ML Model Repositories: The Next Big Supply Chain Attack Target

Researchers warn that public ML model repositories (e.g., Hugging Face) are an emerging supply-chain attack surface where adversaries can distribute malicious models or backdoors—via namesquatting, typosquatting, malicious dependencies, or unsafe serialization formats like pickle—allowing arbitrary code execution and potential full system compromise; real-world examples and demonstrations (JFrog, HiddenLayer, Dropbox research) show these techniques are feasible and underlines the need for stronger model inspection and sandboxing.