AI needs transparency: How supply chain security tools can protect ML models

The report argues that applying software supply chain security practices—particularly Sigstore for signing ML models and SLSA for provenance—can improve integrity and trust in AI systems, but are insufficient on their own. Experts highlight limitations around key management, scope, and the difficulty of assuring training and dynamic data, warning against a false sense of security from signatures alone. The piece advocates extending SLSA with ML-specific metadata (e.g., ML-BOM), adopting more modular and adaptive controls, and increasing vendor transparency to address the unique risks of ML and LLM supply chains.