A central catalog of trained models that tracks every version, along with the metadata and approvals needed to move it into production.
A Model Registry works like a Git repository for AI models. It records the exact datasets, hyperparameter configurations, and evaluation metrics behind every version of a model. That version history makes audits and rollbacks possible.
See also Air-Gapped Environment, Role-Based Access Control and Zero Trust Architecture.