Cisco Launches AI Provenance Tool for Security and Supply Chain Protection

Cisco released the open-source Model Provenance Kit, a Python toolkit and CLI that verifies AI model origins by comparing architecture metadata and performing deeper weight-level analyses (embedding patterns, distribution of learned values, normalization fingerprints, layer energy, and direct weight similarity). The kit supports compare and scan modes, includes a fingerprint dataset for ~150 base models, was evaluated across 111 model pairs with strong detection results, and is intended to detect tampering/backdoors, support compliance, and improve AI supply-chain trust.