The NAIC Is Building a Tool to Grade Your AI Governance. Here's How to Be Ready.

The NAIC's Big Data and AI Working Group is piloting an AI Systems Evaluation Tool for insurers. It's voluntary today, but here's how an agency documents governance before that changes.

Key takeaway

The NAIC’s Big Data and Artificial Intelligence (H) Working Group is piloting an AI Systems Evaluation Tool, an instrument regulators would use to assess how an insurer’s AI is governed. It’s been running in 12 states since early 2026, with wider adoption anticipated at the 2026 Fall National Meeting, and it sits on top of an AI Model Bulletin already adopted by 24 states and D.C. Today it’s voluntary, not a fine and not a mandatory exam, which is exactly the window an agency has to build its AI governance record before answering to it becomes urgent.

What is the NAIC’s AI Systems Evaluation Tool?

It’s an instrument the NAIC’s Big Data and Artificial Intelligence (H) Working Group is building so regulators can assess how an insurer uses AI and machine learning, and how well it governs that use.

The tool has been in a 12-state pilot since early 2026, and wider adoption is anticipated at the 2026 Fall National Meeting. It’s not a new law. It’s a practical, structured way for a state regulator to ask an agency to show its work on AI, the same way an exam already asks an agency to show its work on underwriting files or claims handling. If your agency is already thinking about how AI fits into day-to-day operations, this is the regulatory side of that same question.

Is this tool a fine or a mandatory exam?

No, not yet. It’s currently a voluntary pilot evaluation, not a penalty and not a required exam.

The honest framing matters here. “Wider adoption is anticipated” is a real signal from the working group, but anticipated isn’t certain, and voluntary isn’t mandatory. What is certain is the direction: 24 states and D.C. have already adopted the underlying NAIC AI Model Bulletin that this tool is built to assess against, so the governance expectations behind it are already in place in most of the country. The evaluation tool is the mechanism catching up to those expectations, not a new standard appearing out of nowhere.

What does the tool actually assess?

It assesses governance over AI that touches decisions made on policyholder data, specifically underwriting, claims, and fraud detection.

That framing is worth sitting with. This isn’t a general audit of whether an agency uses AI. It’s specifically about the decisions AI influences for real customers: who gets a policy, at what rate, whether a claim gets paid, and whether a case gets flagged as fraud. Those are the decisions a regulator has always cared about; AI just changes who, or what, is making the first pass at them.

How should an agency get ready now?

Build an inventory: every AI system that touches a customer decision, its data sources, its bias and fairness testing, the human who reviews adverse outcomes, and the date it was last reviewed.

That’s a short list, but most agencies haven’t written it down anywhere a regulator could ask to see it. Agencies that document this ahead of time answer a question like this in an afternoon. Agencies that haven’t spend that afternoon, and several more after it, reconstructing what a system does and who’s responsible for it, usually under a deadline they didn’t set. The inventory itself doesn’t need to be complicated. It needs to exist, and it needs an owner who keeps it current as systems change.

Who owns this work at an agency without a CTO?

A fractional AI officer is the practical answer for agencies that don’t have a full-time technical executive to build and maintain this inventory.

Most independent insurance agencies were never going to hire a full-time chief technology officer to own AI governance, and they shouldn’t need to. That inventory, the data sources, the bias testing, the human reviewer, the review dates, is exactly the kind of ongoing work a fractional AI CTO owns for firms that need the discipline without the headcount. If your agency touches underwriting, claims, or fraud decisions with any AI tool today, that’s worth mapping before the fall, not after. Our services start with exactly that assessment, free.

Frequently asked questions

What is the NAIC's AI Systems Evaluation Tool?

It's an instrument the NAIC's Big Data and Artificial Intelligence (H) Working Group is developing so regulators can assess how an insurer uses AI and machine learning and how well it governs that use. It has been running as a 12-state pilot since early 2026, with wider adoption anticipated at the 2026 Fall National Meeting.

Is the evaluation tool a fine or a mandatory exam?

Not today. It's currently a voluntary pilot evaluation, not a penalty and not a required exam. Wider adoption this fall is anticipated by the NAIC working group, not guaranteed, so the honest read is that this is moving toward standard practice, not that it already is one.

How many states have adopted the underlying AI Model Bulletin?

24 states plus the District of Columbia have adopted the NAIC AI Model Bulletin that the evaluation tool is built to assess against, according to the NAIC Big Data and Artificial Intelligence (H) Working Group.

What does the tool actually look at?

Governance over how AI touches decisions made on policyholder data, specifically underwriting, claims, and fraud detection. That includes what data feeds the model, whether it's been tested for bias, who reviews adverse decisions, and how recently the system was reviewed.

How should an insurance agency get ready?

Build an inventory now: every AI system that touches a customer decision, its data sources, its bias and fairness testing, the human who reviews adverse outcomes, and the date it was last reviewed. Agencies that document this ahead of time answer a regulator's questions in an afternoon; agencies that don't reconstruct it under deadline.