Model Provenance Kit: Know Where Your AI Models Come From (blogs.cisco.com)

Cisco has announced the release of the Model Provenance Kit, an open-source toolkit designed to enhance transparency in AI model lineage. This innovative tool aims to address the pressing issue of model provenance, which is critical for ensuring AI supply chain security. By analyzing both metadata and the inherent learned parameters of AI models, the Model Provenance Kit functions like a "DNA test" for models. It assesses whether different models share a common origin and detects any modifications, helping organizations verify the security, compliance, and ethical implications of their AI systems. The significance of this tool lies in its potential to mitigate risks associated with using AI models from diverse and often unverified sources, such as those available on open repositories like Hugging Face. The toolkit analyzes models through a tiered strategy that starts with fast structural checks and includes deeper weight-level analysis if necessary. Its dual modes allow users to compare two models for lineage or scan a single model against a database of known fingerprints. With its comprehensive approach, the Model Provenance Kit offers organizations a reliable means to track model origins and ensure that the AI they deploy is safe and compliant, ultimately enhancing trust in AI applications across various sectors.