An Empirical Study on Remote Code Execution in Machine Learning Model Hosting Ecosystems
A cross-platform study of model-loading risk across five ML hosting ecosystems, using static analysis, malware-signature scanning, and analysis of more than 600 developer discussions.
ML ecosystem security · public artifact
- Status: Under review at ICSE 2027
- Public boundary: Public arXiv artifact
- Methods: static analysis · CodeQL · Semgrep · YARA · qualitative analysis
- Last verified: 2026-07-10
The study examines how custom code and artifacts enter model-hosting workflows, then maps recurring technical risks and developer misconceptions without treating model popularity as a proxy for safety.