With the NAIC model bulletin adopted across half the states and an AI Systems Evaluation Tool piloted with state regulators, insurers face a concrete artifact list. Here is the checklist, with the continuous-generation approach for each item.
The written AIS program
Governance structure, accountable owners, risk management process, and internal audit hooks for AI systems. Continuous approach: the program document points at living systems — the inventory, the policy set, the evidence stream — rather than describing intentions.
The system inventory
Every AI system touching underwriting, pricing, claims, fraud, or servicing, with third-party systems included. Continuous approach: gateway discovery keeps the inventory current as a side effect of traffic.
Model cards & documentation
Purpose, data, limitations, and performance per system. Continuous approach: cards assembled from registry metadata and live monitoring rather than written once and aging.
Testing records
Bias and outcome testing appropriate to the system’s use, retained with methodology. Continuous approach: scheduled evaluation runs whose results file into the evidence layer automatically.
Drift & performance monitoring
Evidence that deployed systems are watched, with thresholds and escalation. Continuous approach: behavioral baselines per system, alerts tied to autonomy tier.
Consumer-impact traceability
The ability to explain an adverse or unusual outcome for a specific policyholder. Continuous approach: per-decision reconstruction — the same capability your market conduct exam will eventually ask for by name.
Informational only; not legal advice. Confirm requirements per adopting state.