What governance artifacts should be in place for AI models?

• A. Model registry, versioning, data lineage, evaluation reports, and approval workflows.
• B. Customer testimonials.
• C. Only model code.
• D. Marketing plan, budget approvals, vendor contracts, and press releases.

Answer: (A)

Governance of AI models relies on artifacts that document the model’s lifecycle—from data sources to deployment. A model registry acts as a centralized catalog that stores each model version with metadata, status, and lineage, so you can see what’s in production and what’s been retired. Versioning is essential for reproducibility; it records every change to the model so past behavior can be recreated or a safe rollback can be performed. Data lineage traces the journey of data from source through preprocessing to training inputs, helping you assess data quality, detect biases, and demonstrate compliance. Evaluation reports bring together performance metrics, robustness checks, and fairness/safety assessments, informing whether a model meets guardrails before deployment and how it should be monitored afterward. Approval workflows formalize who signs off on changes and when, ensuring governance controls are consistently applied before production.

Taken together, these artifacts provide traceability, accountability, and safer deployment. Other options miss the governance foundation: customer testimonials relate to marketing, not model governance; only model code lacks the metadata and controls needed for reproducibility and auditability; and marketing plans or contracts have no role in managing a model’s risk and lifecycle.