North Carolina Now Expects Insurance Agencies to Document How They Use AI
If your agency uses any kind of AI tool, a chatbot on your website, a quoting assistant, an app that drafts renewal emails, North Carolina's insurance department already has an opinion about how you should be running it. It is not a proposal. It took effect December 18, 2024.
What actually changed
The North Carolina Department of Insurance adopted Bulletin No. 24-B-19, its version of a model bulletin the National Association of Insurance Commissioners (NAIC) first approved in December 2023. As of the NAIC's own tracking, 26 states and jurisdictions have now adopted it in some form, and more are expected to follow at the NAIC's fall meeting this year.
The bulletin's language is aimed at "insurers," and it is easy to read that and assume it does not touch an independent agency. It is not that simple. NCDOI's bulletin says decisions or actions affecting consumers that are made or supported by AI systems still have to comply with the state's existing unfair trade practices and anti-discrimination rules, no matter what technology produced them. If your agency is using AI in underwriting support, quoting, claims triage, or customer communication, that is exactly the kind of consumer-facing decision the bulletin is pointing at.
What the state actually wants you to be able to show
The bulletin does not ban AI. It sets expectations for how it should be governed, and it tells insurers what the department may ask to see if it opens an investigation or exam. In practice, agencies and carriers adopting the model bulletin's framework are expected to have:
- An inventory of the AI systems actually in use, not a vague sense of "we use some AI tools"
- A written program describing who is accountable for each tool and how it gets reviewed
- Documented testing or monitoring for bias and errors, especially anything that touches pricing or eligibility
- A process for evaluating third-party AI tools before they go live, not just after a customer complains
None of that is exotic. It is closer to the paperwork a firm already keeps for a vendor contract or an E&O file. The difference is that most agencies have never been asked to produce it for a chatbot or a quoting tool, so it does not exist yet.
The part agencies miss: buying the tool does not cover you
The detail that catches agencies off guard is accountability. Under the model bulletin's framework, using a vendor's AI product does not transfer responsibility to the vendor. The agency is still expected to be able to show it evaluated the tool, understands roughly how it behaves, and is watching for problems, on top of whatever the vendor's own documentation says. "Our software vendor handles that" is not an answer an examiner is looking for.
That is the same reason we build AI agents for agencies with a human step built in, not a black box that quotes, replies, or files something on its own. When an agent hits something unusual, a person sees it before it goes out. That is not just a customer-service choice. It is the shape of the record a bulletin like this one is asking every agency to be able to produce: who is watching the AI, and what happens when it gets something wrong.
What to do about it now
You do not need outside compliance counsel to take the first step (though for a full read of how the bulletin applies to your specific book of business, that is worth a real conversation with one). Start with a plain list: every place AI touches a customer interaction at your agency, who owns it, and what a person actually checks before anything goes out the door. If that list does not exist yet, that is the gap the bulletin is pointing at, and it is a lot easier to build before an exam than during one.
Sources: NC Department of Insurance, Bulletin No. 24-B-19; NAIC, Implementation of Model Bulletin: Use of Artificial Intelligence Systems by Insurers (status as of August 31, 2026).