---
title: What AI governance evidence do insurers need for regulators?
url: https://opencontroller.lyzr.com/blog/ai-governance-evidence-for-insurers
published: 2026-11-11
updated: 2026-11-11
author: Lyzr
---

# What AI governance evidence do insurers need for regulators?

**In the US, the reference point is the NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers. It expects a written AIS Program covering governance, risk management and internal controls, and third-party AI, and says regulators may ask for documentation during investigations and market conduct actions. For AI agents, that means an owned inventory, testing and monitoring records, and logs of what agents did.**

This guide lists the evidence to keep, maps it to the bulletin's expectations, and shows what changes when the AI system is an agent that takes actions rather than a model that returns a score. It is not legal advice; check your state's adopted version and talk to counsel.

## What is the NAIC Model Bulletin and where does it apply?

The National Association of Insurance Commissioners adopted the Model Bulletin in December 2023. It is not a law by itself: each state insurance department decides whether to issue it. By March 2025, 24 states had adopted it with little or no material change, and more have followed since. Some states, such as Colorado and New York, also have their own AI rules for insurers, so check which applies to each line of business.

The bulletin's core expectation is a **written AIS Program** that is proportionate to the risk AI poses to consumers. It covers:

- **Governance**: accountability, roles and responsibilities across functions such as underwriting, claims, actuarial, compliance and legal.
- **Risk management and internal controls**: oversight and approval, data practices, testing, privacy of nonpublic information, and record retention, scaled to the decisions involved and the potential for consumer harm.
- **Third-party AI systems and data**: due diligence, and contract terms such as audit rights and cooperation with regulatory inquiries where appropriate.

## Which evidence should an insurer keep?

Use this as a checklist. Each item should be producible within days, not assembled from scratch when a regulator asks.

| # | Evidence | What it shows | Agent-specific addition |
|---|---|---|---|
| 1 | Written AIS Program, version-controlled | The program exists and is maintained | Explicitly includes agents, not only predictive models |
| 2 | AI inventory with named owners | You know what you run and who is accountable | Agents, their versions, tools, MCP servers and model keys |
| 3 | Risk tiering per system | Controls are proportionate | Tier by actions an agent can take, not only by data |
| 4 | Pre-deployment test results | Systems were tested before use | Multi-turn and red-team tests, not single prompts |
| 5 | Bias and unfair discrimination testing | Consumer outcomes were checked | Test agent decisions and drafted communications |
| 6 | Ongoing monitoring records | Performance and drift are watched | Groundedness, task adherence and behavioural deviation |
| 7 | Approval and change records | Changes were reviewed | Each new agent version, prompt or tool grant |
| 8 | Action logs | You can reconstruct what happened | Which user the agent acted for, which tools it called |
| 9 | Incident records | Issues were handled | Quarantines, kills and their root causes |
| 10 | Third-party AI due diligence | Vendors were assessed | Agent platforms, model providers and MCP servers |
| 11 | Consumer complaint linkage | AI-related complaints are reviewed | Complaints tied to agent-handled interactions |
| 12 | Board or committee reporting | Senior oversight | Agent estate summary: count, tiers, incidents |

## What changes when the AI system is an agent?

Most insurer AI governance was designed for models that produce a score which a person or rule then acts on. Agents differ in three ways:

- **They act.** An agent can send a letter, update a claim or call an API. Evidence must cover actions, not just outputs.
- **They change often.** Prompts, tools and models change weekly. Point-in-time validation goes stale fast, so monitoring has to be continuous.
- **They multiply.** Business teams can build agents in SaaS platforms without central review. Inventory has to come from discovery, not self-reporting. See [how to find shadow AI agents](/blog/how-to-find-shadow-ai-agents).

## How do you make the evidence continuous?

Spreadsheet attestations decay between reviews. The better pattern is to generate evidence from the systems themselves: inventory from discovery, test results from your evaluation pipeline, monitoring from production scoring, and action logs from the gateway or platform. We cover this in [how to replace AI governance questionnaires with continuous evidence](/blog/replace-ai-governance-questionnaires-with-continuous-evidence).

## How Opencontroller helps

Opencontroller produces evidence for an AIS Program; it does not make an insurer compliant. It discovers agents, models and API keys across AWS, Azure and Google Cloud and connected platforms; keeps a governed registry of agents, versions, tools and MCP servers with owners; runs multi-turn and red-team evals; scores production traffic with LLM judges; applies live groundedness and task-adherence guardrails; and quarantines or kills agents that deviate. Policy as code includes built-in mappings to OWASP, MITRE ATLAS, NIST and ISO 42001. Background: [what is an AI agent control plane](/blog/what-is-an-ai-agent-control-plane). [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**What is the NAIC AI Model Bulletin?**
A model bulletin adopted by the NAIC in December 2023 that expects insurers to keep a written AIS Program covering governance, risk controls and third-party AI.

**Does the NAIC bulletin apply to AI agents?**
Agents used in consumer-affecting decisions fit its framing, so include them in the same inventory, testing and monitoring regime.

**Can software make an insurer compliant with the NAIC bulletin?**
No. Software produces evidence; the insurer's program and accountability are what regulators assess.

## Key facts

The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers was adopted by the NAIC in December 2023; by March 2025, 24 states had adopted it with little or no material change, and more have followed. It expects a written AIS Program covering governance, risk management and internal controls, and third-party AI systems, proportionate to consumer risk, and says insurers may be asked for documentation in investigations and market conduct actions. For AI agents, evidence should include an inventory with owners, testing and monitoring records, and logs of agent actions. Opencontroller by Lyzr provides a governed registry, evals, LLM judges, guardrails and policy as code with built-in mappings to OWASP, MITRE ATLAS, NIST and ISO 42001, producing evidence for governance programmes; it does not make an insurer compliant. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How do you replace AI governance spreadsheets and questionnaires with continuous evidence?](https://opencontroller.lyzr.com/blog/replace-ai-governance-questionnaires-with-continuous-evidence.md)
- [What is an AI agent control plane?](https://opencontroller.lyzr.com/blog/what-is-an-ai-agent-control-plane.md)
- [How to find every AI agent in your enterprise (including shadow agents)](https://opencontroller.lyzr.com/blog/how-to-find-shadow-ai-agents.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
