Governance

Model Registry

A database or system for recording, tracking, and managing AI models throughout their lifecycle.

Definition

Model Registry refers to systems for cataloging, versioning, and tracking AI models throughout their lifecycle—from development through deployment and retirement. It serves both internal governance and regulatory compliance purposes.

EU AI Act Requirements: The Act mandates registration in an EU-level database:

  • Article 71: Establishes EU database for high-risk AI systems
  • Registration required: Providers must register high-risk systems before placing on market
  • Information included: Provider details, system description, conformity status, intended purpose
  • Public access: Much information publicly searchable

Internal Registry Functions:

  • Version control: Track model iterations and changes
  • Metadata management: Training data, parameters, performance metrics
  • Lineage tracking: Document model provenance and dependencies
  • Deployment tracking: Where and how models are deployed
  • Access control: Manage who can use or modify models
  • Audit trail: Record all model-related activities

Regulatory Benefits:

  • Demonstrates governance and oversight
  • Facilitates incident investigation
  • Supports conformity assessment
  • Enables post-market monitoring

Tools: MLflow Model Registry, AWS SageMaker Model Registry, Azure ML Model Registry, custom enterprise solutions.

Related concepts: EU Database for High-Risk AI Systems, Technical Documentation, Post-Market Monitoring, Audit Trail

Sources

  • SR 11-7
  • Model Risk Management Best Practices