AI Governance Checklist

Use this checklist before deploying any AI system and as part of regular governance reviews of existing deployments. Each item is a concrete, verifiable governance criterion.

Before Deployment

• We have defined what decision or process this AI system is intended to support
• We have identified a named individual accountable for this system's performance
• We have assessed the quality, representativeness and recency of the data the system uses
• We have conducted a pre-deployment risk assessment identifying potential harms
• We have defined what human review process exists for high-stakes AI-assisted decisions

Transparency and Explainability

• People affected by AI-assisted decisions can be informed that AI was involved
• The system can provide a meaningful explanation for specific outputs when required
• We have documented what the system does and does not do in plain language
• Staff who use the system have received training on its capabilities and limitations
• We maintain an audit log of decisions made with AI assistance in high-stakes contexts

Ongoing Monitoring

• We have defined performance metrics and thresholds that trigger a review
• We monitor model performance on an ongoing basis (not just at deployment)
• We have a process for investigating when the system produces unexpected or harmful outputs
• We review performance disaggregated by relevant subgroups (not just overall accuracy)
• We have a defined schedule for model review and re-validation

Vendor and Procurement

• We have asked vendors how their models are trained, validated and monitored
• We have confirmed what data the vendor uses to train models and whether it includes our data
• Our contract with the vendor includes provisions for performance accountability
• We know how the vendor handles model incidents and what remedies exist
• We have assessed whether switching vendors would result in loss of model history or data