Cisco Open-Sources Model Provenance Kit to Secure AI Supply Chains

Companies are increasingly deploying third-party and open-source AI models without a reliable way to verify how they were built or whether they were derived from another system. Cisco’s Model Provenance Kit is designed to close that transparency gap, helping organizations assess copyright, licensing and regulatory exposure while strengthening security controls across an AI supply chain that can include multiple model developers and distributors.

Cisco has open-sourced the Model Provenance Kit, giving enterprises a tool to investigate the origin and lineage of external AI models. The software examines architecture metadata and signals embedded in model weights to determine whether a model was trained independently or derived from another model through processes such as fine-tuning. Cisco says the approach can accurately distinguish model relationships, enabling security and compliance teams to flag unidentified or potentially misrepresented models before deployment.