What Washington Expects Before an Insurer Deploys AI
Washington's insurance regulator treats a written AI governance program as the expected way to show that an AI-driven underwriting, pricing, or claims decision still meets state law. Here is what to put in place before you deploy.
In this guide
Before a Washington insurer or managing general agent turns on an AI model that helps decide who gets a policy, what they pay, or how a claim is handled, the question to settle is not whether the tool works. It is whether you can show the state that the decision it supports still meets Washington insurance law. The short version: Washington has not passed a binding AI insurance rule, but its Office of the Insurance Commissioner issued a 2024 advisory that treats a written AI governance program as the expected way to demonstrate that, and the unfair-practices and anti-discrimination laws behind it apply no matter which tool made the call.
Does Washington require insurers to have an AI program?
No single rule compels it, but the expectation is on the record. In April 2024 the Washington State Office of the Insurance Commissioner issued Technical Assistance Advisory 2024-02, The Use of Artificial Intelligence Systems in Insurance, addressed to all insurers that hold a certificate of authority to do business in the state. The advisory describes itself as advisory only, a policy statement under RCW 34.05.230(1) rather than a new regulation. What it is not optional about is the law it rests on: the Unfair Trade Practices Act (Chapter 48.30 RCW), the unfair-discrimination statute (RCW 48.18.480), the unfair claims settlement rules (WAC 284-30-300 through 284-30-390), and property-and-casualty rating standards (RCW 48.19.020) all apply to a decision regardless of the method used to reach it. An AI model does not create a safe harbor; it only changes how you have to document that the outcome is fair.
The advisory's answer to how you show that is a written program it calls an AIS Program. Washington is one of many states that adopted the model bulletin the National Association of Insurance Commissioners released in December 2023, which is a guiding document rather than a model law. In 2025 Commissioner Patty Kuderer went further and convened an AI Advisory Board, which held its first meeting in Seattle on June 30, 2025; the board studies how insurers use AI and advises the office, but it has no regulatory or enforcement authority of its own. The direction is clear even though the current instrument is guidance.
See how Clairvance builds AI governance into an insurance implementation
Who the advisory covers, and why an MGA should still read it
It binds licensed insurers directly, but its reach extends by contract. The advisory is addressed to carriers with a certificate of authority, so a retail agency that uses AI only to draft emails or route service requests is not its direct subject, though that agency is still bound by the same unfair-practices law. The entity most exposed after the carrier is the managing general agent or program administrator. If you make or support underwriting or claims decisions under delegated authority, the advisory treats you as the third party a carrier must perform due diligence on, and the carrier is told to confirm that your AI will meet the legal standards imposed on the insurer itself. In practice the carrier's obligations land in your contract, so the cheapest time to build the program is before a carrier audit asks for it.
What a written AIS Program is supposed to cover
At a minimum, a program addressing governance, risk management controls, and internal audit, owned by someone senior who answers to the board. The advisory does not hand you a template. It describes a program designed to mitigate the risk of adverse consumer outcomes, with responsibility for it vested in senior management accountable to the board. It expects the program to follow an AI system across its whole life cycle, from design and validation through deployment, monitoring, and retirement, and across the insurance life cycle, from product design and marketing through underwriting, rating, claims, and fraud detection. It also expects you to tell affected consumers that AI systems are in use.
The part that keeps the program from becoming a binder nobody opens is the advisory's own instruction that controls be commensurate with the risk of a given use. It names five things to weigh: the nature of the decision the model informs; the type and degree of potential harm to the consumer; how much a human is involved in the final decision; how transparent and explainable the outcome is to the person affected; and how much you rely on third-party data and models. Those five factors are the useful core, because they let you spend governance where harm is real instead of spreading it evenly across every use.
Sort your AI uses by consumer harm before you write a line of policy
A practical way to apply those five factors is to inventory every place AI touches a decision and tier each one by how badly a wrong answer hurts a consumer and how much a human still decides. This tiering is a planning method, not legal advice or an official Washington requirement, but it maps directly onto the advisory's commensurate-with-risk language and shows you where to start. The three uses below are illustrative examples, not a Clairvance client or a Washington ruling.
| AI use (illustrative) | Consumer harm if wrong | Human role | Minimum controls to attach first |
|---|---|---|---|
| Generative tool drafting renewal reminder emails | Low: no coverage or price decision | Staff sends; AI only drafts | Light: a usage log, a human send step, and periodic spot-checks |
| Model flagging claims for special investigation | Medium: delay and added scrutiny, not denial | Adjuster decides; the flag is advisory | Bias testing on the flags, documented thresholds, and a record of adjuster overrides |
| Model scoring underwriting eligibility or price | High: it sets who is covered and at what cost | Model output drives the decision | Full: pre-deployment and ongoing discrimination testing, model documentation, drift monitoring, a named accountable owner, and consumer notice |
The point of the tiers is triage. A renewal-email drafter does not need the same file as a pricing model, and pretending it does is how a program stalls before it protects anyone. Start the written program with the high-harm row, because that is the use an examiner will ask about first and the one where an unfair-discrimination finding actually bites.
If you buy or delegate the model, the contract carries the rules
Do due diligence on the vendor, then put audit and cooperation rights in the contract. Most insurers will not build these models; they will license them or inherit them through an MGA. The advisory anticipates that and devotes a section to third-party systems. It expects you to assess the third party well enough to be confident its outputs will meet the legal standards imposed on the insurer itself, and, where appropriate and available, to include contract terms that give you audit rights or audit reports and require the vendor to cooperate with regulatory inquiries and investigations. If a vendor will not let you or a regulator see how the model behaves, that is a decision to make before signing, not after an examination opens. Keep the due-diligence record you used to choose the vendor, because it is part of what the office can ask about later.
What an examiner can ask for later
The written program itself, proof you adopted it, its scope, and your model documentation. Section 4 of the advisory lists what the office may request in an investigation or market conduct action: the written AIS Program, information evidencing its adoption, the scope of what it covers, including any AI systems it leaves out, and documentation about specific models. Two practical consequences follow. First, good intentions are not what gets requested; a document is. Second, the scope question means you should write down the uses you deliberately excluded and why, rather than leaving gaps an examiner finds for you. Build the file as you deploy, because reconstructing it a year later from memory is the expensive path.
None of this is a reason to avoid AI in insurance; the office says plainly that it encourages innovation that contributes to safe and stable markets. It is a reason to treat governance as part of the implementation rather than a compliance chore bolted on afterward. For the rules that apply to any Washington business turning on an AI tool, such as call-recording consent, health-data limits, and breach notice, see our guide on what Washington law requires before you turn on an AI tool; this article is the narrower insurance-regulator layer that sits on top of those.
Quick answers
Is Washington's insurance AI advisory a law you can be fined for breaking?
The advisory itself is a policy statement issued under RCW 34.05.230(1) and is advisory only. But it rests on binding laws, including the Unfair Trade Practices Act, the unfair-discrimination statute, and the claims-settlement and rating rules, that apply to an AI-supported decision regardless of the method used to reach it.
Does a small insurance agency need an AIS Program?
The advisory is addressed to licensed insurers holding a certificate of authority. An agency using AI only for internal drafting or routing is not its direct subject, though it is still bound by unfair-practices law. A managing general agent with delegated underwriting or claims authority should expect the program expectations to reach it through its carrier contract.
Where should an insurer start?
Inventory where AI touches a consumer decision, tier each use by potential harm and human involvement, and build the written program around the highest-harm use first, typically underwriting eligibility, pricing, or a claim denial, with discrimination testing, model documentation, and a named accountable owner.
What belongs in a contract for a third-party AI model?
The advisory points to audit rights or audit reports and a requirement that the vendor cooperate with regulatory inquiries. Keep the due-diligence record you used to select the vendor as well, since the office can ask how you assessed a third-party system.
Sources
- Washington State Office of the Insurance Commissioner: Technical Assistance Advisory 2024-02, The Use of Artificial Intelligence Systems in Insurance (April 22, 2024)
- Washington State Office of the Insurance Commissioner: Kuderer convenes advisory board on insurers' use of AI (July 1, 2025)
- The National Law Review: State Regulators Address Insurers' Use of AI, 11 States Adopt NAIC Model Bulletin (June 7, 2024)
How we research and review these guides
Put the ideas to work in your business.
We provide AI consulting and implementation for insurance agencies and brokers. Bring us the process that is slowing your team down and the systems involved. We can assess the problem with you and discuss a practical implementation.
Discuss your project →Still exploring? Explore the Workflow Opportunity Workbook →