Property insurance underwriting has always been about information. The insurer wants to know the condition of the property, its location, its hazards, its construction, its protection, its maintenance, and its likely chance of loss. What is new is the extraordinary amount of information now being gathered about policyholders and their property without anybody knocking on the door, climbing on the roof, walking the site, or even telling the policyholder what is being collected.
Artificial intelligence has changed the underwriting conversation. In a recent Digital Insurance article, “How fairness must be central in underwriting,” 1 I noted that AI can now find, organize, interpret, and flag property information before underwriting even begins. Aerial imagery, satellite data, permit records, tax records, wildfire models, flood maps, roof analytics, claim history, vegetation measurements, code enforcement records, and third-party risk scores can now be pulled together into a digital underwriting file before the policyholder has any meaningful opportunity to explain the truth about the property.
The issue is that the computer may find data, but it does not always find the truth. A roof may look stained from an aerial photograph but be structurally sound. A shadow may look like deterioration. A permit record may be incomplete. A wildfire score may not reflect recent defensible space work. A building department file may lag behind completed repairs. A condominium association may have spent millions on mitigation, fire safety improvements, roof work, drainage improvements, or code upgrades that an algorithm never recognizes. The machine may be fast, but speed is not fairness.
The property insurance industry sometimes talks about these tools as if a neutral “AI risk score” is simply producing an objective underwriting answer. This is often an oversimplified view. In many instances, AI is not the final decision-maker. It is the scout, the file clerk, the investigator, and sometimes the final judge about acceptance, rate, and price. It gathers information, sorts it, ranks it, flags it, and presents it to an underwriting system which is also robotic. The insurer’s own underwriting rules, sometimes based on algorithms, decide whether the policyholder pays more, receives less, is forced to repair, is restricted, is non-renewed, or cannot buy coverage at all.
A third-party vendor roof algorithm may say the roof appears worn. A wildfire model may say the property lacks sufficient defensible space. A computer vision program may identify trees, debris, solar panels, a pool, a trampoline, roof discoloration, or neighboring exposures. But the insurance consequence happens when the carrier applies its own rulebook to this information. One insurer may tolerate the condition. Another may surcharge it. Another may exclude coverage. Another may refuse the risk altogether.
Policyholders are not just being scored. They are being judged under rules they often cannot see. This is unacceptable in a regulated industry built on public trust.
The California Department of Insurance has already recognized a central truth that deserves far more attention. Underwriting rules submitted by insurers in connection with rate applications are public information. They are not secret simply because an insurer stamps them “confidential,” “proprietary,” or “trade secret.” Underwriting rules are not academic paperwork. They are the operating instructions that determine which consumers are accepted, rejected, classified, surcharged, restricted, or non-renewed.
Those rules should be public because they affect the public. Insurance is not a private club selling luxury goods to people who can take it or leave it. Property insurance is a public product to serve the public. It is essential to homeownership, lending, business survival, condominium governance, and community recovery after catastrophe. In many places, citizens cannot meaningfully participate in the economy without it. When insurers are given the privilege of selling this essential financial product in a regulated marketplace, they cannot be allowed to hide the rules of the game.
The insurance industry will often respond that underwriting rules, models, and vendor data involve trade secrets. Some of that concern may be legitimate. Nobody is suggesting that every line of software code or every proprietary business method must be casually posted on the courthouse wall. But trade secret protection cannot become regulatory immunity. It cannot be the magic phrase that ends oversight. It cannot be used as a curtain behind which unfair discrimination, inaccurate data, unjustified nonrenewals, or market withdrawals are hidden from regulators and consumers. A referee cannot call a fair game if one team is allowed to hide the rulebook.
The problem becomes even more serious as coverage shrinks while premiums rise. Policyholders across the country are paying more and receiving less. Deductibles are increasing. Roof coverage is being limited. Water damage coverage is being narrowed. Managed repair provisions are spreading. Wildfire, hail, wind, and flood exposures are being carved, capped, limited, or priced beyond reach. Consumers are already dealing with a marketplace that too often feels like a take-it-or-leave-it proposition. Adding invisible AI-assisted underwriting to that reality only deepens the imbalance and invites public distrust of the insurance product.
Good underwriting requires good information. Better data can help insurers price more accurately, reward mitigation, distinguish one property from another, and avoid broad-brush decisions that punish entire neighborhoods or counties. Nobody wants underwriting based on ignorance. But there is a world of difference between informed underwriting and secret underwriting.
Fair underwriting in the age of artificial intelligence should require three basic principles. The policyholder should be told what specific property characteristics drove the decision. The policyholder should be allowed to see and correct the information being relied upon. The insurer should be able to explain the rule that turned that information into a premium increase, repair demand, restriction, cancellation, or nonrenewal.
Regulators should demand that insurers using AI-assisted underwriting maintain governance programs, validate data quality, test for unfair discrimination, supervise third-party vendors, and document how adverse underwriting decisions are made. Insurers should not be permitted to outsource accountability to a vendor. If the insurer uses the score, image, model, or data point, the insurer owns the consequence.
Policyholders should also be given a meaningful opportunity to respond. Not a hollow form letter or a vague notice saying the property no longer meets “underwriting guidelines.” What image was used? What date was it taken? What condition was identified? What rule was applied? What can the policyholder submit to correct the record? What mitigation will be credited? Who will review the challenge? Will a human being actually look at the evidence?
These questions matter because underwriting decisions increasingly determine whether families can keep their homes insured, whether condominium associations can obtain coverage, whether businesses can satisfy loan requirements, and whether communities in catastrophe-prone areas remain economically viable.
Artificial intelligence will remain part of property insurance underwriting. It should. Used properly, it can help identify risks, encourage mitigation, reduce uncertainty, and improve pricing. But AI must serve the promise of insurance rather than undermine it. The promise of insurance is not simply to collect premiums when the sun shines and disappear when the model turns red. The promise is financial protection, fair treatment, and good faith. The promise requires transparency.
If an insurer is going to use AI, algorithms, aerial imagery, third-party models, or risk scoring to make decisions affecting premiums, eligibility, renewal, or coverage, the policyholder deserves to know what was used, why it mattered, and how to challenge it. Regulators deserve to see the underwriting rules. The public deserves to understand how essential insurance products are being priced and restricted.
Technology may change the tools of underwriting. It should not change the moral obligation to be fair.
Thought For The Day
“For everything that AI can do, AI can’t decide which problems are worth solving.”
— Lisa Su
1 Chip Merlin. “How Fairness must be central in underwriting.” Digital Insurance. (June 10, 2026). (Available online at https://www.dig-in.com/opinion/how-fairness-must-be-central-in-underwriting)