Insurance AI Control Plane NAIC | Moring
The NAIC sets the baseline expectations for how US insurers govern AI: a written AIS program, lifecycle governance, testing for unfair discrimination, and documentation an examiner can request. An AI control plane is how a carrier makes those expectations operational, turning program requirements into runtime controls and examination-ready evidence.
Short answer
The NAIC sets the baseline expectations for how US insurers govern AI: a written AIS program, lifecycle governance, testing for unfair discrimination, and documentation an examiner can request. An AI control plane is how a carrier makes those expectations operational, turning program requirements into runtime controls and examination-ready evidence.
How it works
The National Association of Insurance Commissioners has shaped how US insurers are expected to govern AI. The anchor document is the Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted in December 2023 and since adopted by a growing share of state jurisdictions, with several large states adding their own insurance-specific regulation. The bulletin does not create new substantive law; it reminds carriers that existing unfair trade practices and unfair discrimination rules apply in full when a decision is made or supported by an AI system.
The bulletin's central expectation is a written AIS Program covering governance, risk management, and internal controls across the AI lifecycle, with accountability reaching board and senior management level. Section 4 adds the examination posture: in an investigation or market conduct action, an insurer can expect questions about how it develops, deploys, and uses AI systems, the outcomes of those systems, and its governance framework, regardless of whether a written program exists. Documentation an examiner can review is the operative standard, not a policy statement of intent.
A control plane is the mechanism that makes those expectations operational. The bulletin expects oversight across development, acquisition, validation, deployment, and ongoing monitoring; the control plane supplies the runtime half, enforcing scope on agent behavior, holding consequential actions for named reviewers, and recording policy decisions, tool calls, approvals, and outcomes in a reviewable trail. The program document and the running system then say the same thing, which is the consistency examiners probe for.
Third-party accountability is a point carriers underestimate. The bulletin makes the insurer responsible for AI sourced from vendors, and adoption data shows the exposure: roughly 40% of models in the NAIC auto survey came from more than 70 third-party vendors. The control plane's registry covers vendor models and tools alongside internal ones, so accountability has a single inventory behind it, and vendor behavior in production is monitored by the same layer that governs internal agents.
Examination readiness is the practical deliverable. When a state department asks how a specific claim decision was reached, the carrier should produce the action record: what the agent proposed, which policy allowed it, what a human approved, and what changed. When it asks how the AIS Program operates, the governance body's decisions, the risk tiers, and the monitoring evidence should be retrievable. A control plane produces both continuously; assembling them during an examination is the failure mode the bulletin's language anticipates.
For program owners, the mapping is direct. The written program defines governance structure, risk tiering, and internal controls; the control plane enforces them at runtime; periodic review uses the runtime record as its evidence base. State-specific requirements sit on top of the NAIC baseline, so carriers should check the map for the jurisdictions where they write business and confirm the control plane's evidence covers what each department asks.
Frequently asked questions
What does the NAIC expect from insurers using AI?
A written AI Systems Program covering governance, risk management, and internal controls across the AI lifecycle, with board and senior management accountability, proportionate controls tied to consumer impact, and documentation an examiner can review. Decisions made or supported by AI must comply with existing insurance law, including unfair trade practices and unfair discrimination rules. In an investigation or market conduct action, insurers can expect questions about how they develop, deploy, and use AI systems and the outcomes those systems produce, regardless of whether a written program exists yet.
How does the NAIC bulletin treat third-party AI used by carriers?
The insurer remains accountable. The bulletin states plainly that responsibility for consumer-impacting decisions does not transfer to the vendor that supplied the model or tool, and adoption data underlines the exposure: about 40% of models in the NAIC's private passenger auto survey came from more than 70 third-party vendors. Practically, this means vendor models and tools belong in the same inventory, monitoring, and evidence trail as internal systems, so the carrier can show how each third-party component was validated, deployed, and supervised in production.
How does an AI control plane support NAIC examination readiness?
By producing the evidence continuously instead of assembling it under deadline. The control plane records what each agent proposed, which policy version allowed the action, what a human approved, and what changed in the source system, keyed to identity and workflow. When an examiner asks how a specific claim or underwriting decision was reached, the carrier retrieves the action record; when asked how the AIS Program operates, the governance decisions, risk tiers, and monitoring evidence are retrievable from the same layer. The program document and the running system say the same thing.
Does the NAIC bulletin apply to agentic systems, not just predictive models?
Yes, because the bulletin's definition of an AI system is broad: a machine-based system that generates outputs such as predictions, recommendations, or decisions influencing insurance-related matters. It reaches across underwriting, rating, claims, marketing, and fraud detection, and an agent that gathers claim evidence, applies appetite rules, or prepares a settlement recommendation falls within it. The bulletin was written before agentic systems were widespread, but its framework, governance, risk management, and internal controls proportionate to consumer impact, applies to them directly, which is what makes runtime controls the practical implementation.