How do you implement model governance, lineage and auditability for regulated ML?

Combine SageMaker lineage tracking, the Model Registry with approval workflow, Model Cards for documented intent and limitations, CloudTrail for attribution, and versioned data snapshots — so any production decision can be traced to the exact model, data, code and approver.

The regulatory question is always a chain: this decision → which model → which artefact → which training job → which data and code → who approved it and on what evidence. Governance means every link is recorded automatically rather than reconstructed later.

The layers

1. Lineage — SageMaker automatically records artefacts (datasets, images, models), actions (jobs) and associations. You can query the graph:…