The NAIC AI Model Bulletin, made practical

A written program with clear ownership. Senior management is accountable to the board, and a cross-functional body oversees AI across its whole life cycle.

• Board-level accountability with named senior-management ownership
• A committee spanning actuarial, data science, underwriting, claims, compliance, and legal
• Documented chains of command, decision rights, and escalation paths
• Ongoing training and supervision for the people who build and use the models

Controls at every stage of the model life cycle, from data sourcing through retirement, sized to the potential harm to consumers.

• An inventory and description of every predictive model and AI system in use
• Data practices covering lineage, quality, integrity, bias analysis, and currency
• Validation and testing that compares development performance against production behavior
• Monitoring for model drift and protection of non-public consumer information

The insurer stays responsible for AI it did not build. Vendor relationships need diligence, contract rights, and the ability to produce evidence.

• Due diligence on vendors, their data, and their models before adoption
• Contract terms granting audit rights and audit-report access
• Cooperation clauses so vendors support regulatory inquiries
• Ongoing confirmation that vendors meet their contractual and regulatory obligations

Section 4 spells out what an examiner can ask for. Treating that list as a standing requirement is what keeps a program defensible.

• The written program, adoption record, and evidence of how it is tailored
• Model documentation: goals, development, validation, and oversight decisions
• Records of data sources, techniques, thresholds, and controls per model
• Validation, testing, and audit records, including model-drift evaluation