Shipped: AI Model Provenance — Hugging Face Blockchain Registry

An open-source AI security proof of concept exploring how blockchain-style recordkeeping can support AI model registration, metadata integrity, and model provenance tracking.

The project retrieves pretrained models from the Hugging Face Model Hub, extracts metadata including model architecture, vocabulary size, special tokens, model classification, and modification timestamps, and records this information in a cryptographically linked blockchain-style registry.

Built in Python and Jupyter Notebook, the framework demonstrates how sequential hashing and structured metadata records can create a tamper-evident history of AI model registrations. Model information is also exported to CSV and blockchain records to JSON for inspection and analysis.

The research explores potential applications in AI supply chain security, model governance, provenance verification, and trustworthy machine learning infrastructure.

This is an educational research prototype demonstrating metadata-level integrity, not a production blockchain network or a complete verification system for neural network weights.