ML Model Signing: Cryptographically Paving the Way to Provenance in Machine Learning Models

Explore the critical issue of machine learning model provenance in this 21-minute conference talk by Mihai Maruseac from Google, presented at OpenSSF. Delve into the challenges of verifying the origin and integrity of ML models, especially as their adoption skyrockets with over 600,000 models available on repositories like Hugging Face. Discover why current cryptographic signing methods fall short for ML models, which consist of multiple files in various formats. Learn about an open-source specification and implementation for cryptographically signing ML model file collections, enabling trust between model producers and end-users. Understand how this approach lays the groundwork for comprehensive model provenance, potentially revealing crucial information about training frameworks, datasets, and model creators, ultimately strengthening the AI supply chain.