A homeowner receives a notice of nonrenewal. The reason is not an inspection by a person. Nobody knocked on the door. Nobody climbed a ladder. Nobody spoke with the homeowner. Instead, an algorithm reviewing aerial imagery flagged the property as having a deteriorated roof, excessive vegetation, or another condition deemed too risky for continued coverage.
Welcome to the new world of property insurance underwriting. Artificial intelligence, aerial imagery, satellite technology, drones, predictive analytics, and massive property databases are rapidly transforming how insurers evaluate risk. What was once science fiction has become standard operating procedure throughout much of the property insurance industry.
According to research conducted through the National Association of Insurance Commissioners (NAIC), more than 70 percent of homeowners insurers reported that they were either using, developing, or exploring artificial intelligence and machine learning systems. 1 Nearly half reported using these tools in underwriting, with many others actively building and testing AI-driven underwriting platforms.
Importantly, that NAIC survey was published nearly three years ago. The findings reflected a marketplace where many insurers were still in the planning, development, or pilot-testing stages of implementation. Given the speed at which artificial intelligence technology has evolved and the significant investments insurers have made since then, it is reasonable to conclude that much of what was then categorized as “planned” or “under development” has now become part of everyday underwriting operations. The reality in 2026 is likely that AI adoption among property insurers is substantially greater than the survey numbers suggest.
Today, these technologies are not merely assisting human underwriters. In many instances, they are becoming the primary gatekeepers, determining whether insurance is offered, renewed, restricted, or declined altogether.
The technology itself is impressive. Companies such as Nearmap, Cotality (formerly CoreLogic), Cape Analytics, Verisk, Zesty.ai, and others provide insurers with extraordinary amounts of information about individual structures. Artificial intelligence can estimate roof age, identify roofing materials, detect tree overhangs, analyze vegetation, estimate replacement costs, evaluate wildfire exposure, and identify maintenance conditions without anyone ever visiting the property. Insurers can combine this information with permit histories, claims histories, weather databases, and public records to create highly detailed risk profiles.
The problem is not necessarily the technology. The problem is what happens when the technology is wrong.
Over the past several years, policyholders have increasingly reported situations where coverage decisions were based upon outdated, inaccurate, misleading, or misunderstood aerial images. Roofs have been identified as damaged when repairs have already been completed. Shadows have been interpreted as defects. Trees have been flagged despite being removed months earlier. Policyholders often receive adverse underwriting decisions without ever seeing the image or data upon which the insurer relied. I have noted this in AI and Drones Are Judging Your Claim and Property Without You Knowing, and “I Was Droned!” California Policyholders Are Being Monitored Regarding Their Loss Risk Exposure and Loss Mitigation Attempts.
That has caught the attention of insurance regulators. The regulatory response is not aimed at prohibiting artificial intelligence. Regulators understand that technological innovation is here to stay. Instead, they are attempting to impose accountability, transparency, and fairness on systems that can otherwise operate as opaque black boxes.
The NAIC adopted its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers and, remarkably for insurance regulation, states have moved quickly to implement it. More than half of the states have adopted the bulletin or substantially similar guidance. The bulletin requires insurers to maintain governance programs for artificial intelligence systems, monitor outcomes, evaluate potential bias, validate data quality, oversee third-party vendors, and ensure that consumer impacts are properly addressed.
Perhaps even more interesting is the growing wave of state-specific regulation focused directly on aerial imagery and AI-assisted underwriting. California has become a leading example. Assembly Bill 1559 would require insurers to notify homeowners when aerial imagery may be collected or used regarding their property. More importantly, if an insurer relies upon aerial imagery to cancel, nonrenew, or reduce coverage, the insurer would be required to provide that image to the policyholder, allow the policyholder to dispute its accuracy, and provide an opportunity to demonstrate remediation before the coverage decision becomes effective. The legislation would also prohibit insurers from relying on aerial images more than 180 days old unless the condition has been independently verified.
This is good legislation and regulation because the technology is not failsafe. People should be allowed to see and challenge the evidence being used against them.
Other states are issuing similar guidance. Regulators increasingly emphasize that insurers cannot simply outsource responsibility to technology vendors. If an artificial intelligence system produces inaccurate results, regulators expect insurers to own those results. The insurer remains responsible for ensuring that the data is current, accurate, relevant, and fairly applied.
The legal question is no longer whether insurers may use artificial intelligence. They already do. The real questions are whether the information is accurate, whether the decision-making process is transparent, whether policyholders are given meaningful due process, and whether insurers can explain and defend the conclusions generated by machines.
At the recent GenRe claims conference, it was refreshing that the claims executives seemed to accept this attitude. They recognize that AI and other technologies are subject to error and that human oversight is important.
Artificial intelligence may analyze thousands of images in seconds. It may process data far faster than any human being. It may even identify risks that human inspectors miss. If an insurer uses artificial intelligence to make decisions affecting homeowners, policyholders must have the right to know what information was used, how it was used, and whether it was accurate. Otherwise, we are simply replacing one form of arbitrary decision-making with a more sophisticated version hidden behind a computer screen.
Thought For The Day
1 “The real problem is not whether machines think but whether men do.”
— B.F. Skinner
Home Insurance Artificial Intelligence/Machine Learning Survey Results, NAIC Staff Report, Aug. 10, 2023.