Linda Bennett has lived in her Santa Ana home since 1993. Last spring, State Farm sent her a letter demanding she replace her roof with approved materials or lose coverage. Cost estimate: roughly $28,000. No inspector had visited the property, and no adjuster had climbed a ladder. Bennett told ABC7 she believes the company evaluated her roof remotely, from aerial imagery she never consented to and never saw, and delivered a verdict that arrived by mail like a parking ticket.
She had until May 1 to comply or find a new insurer. State Farm confirmed that aerial imagery is "one of several tools" it uses when reviewing properties.
Bennett is not an outlier but a data point in an industry-wide shift that has moved faster than the regulation designed to contain it, and it matters to anyone building or buying a home right now because the algorithm scoring your roof does not distinguish between a thirty-year-old asphalt shingle and a two-year-old architectural shingle that a tree shadow made look worse than it is.
How the Scoring Works
CAPE Analytics, a Moody's company, pioneered AI-driven roof condition assessment from aerial imagery in 2016. By February 2025, the company had launched version five of its Roof Condition Rating model, which scores properties across 110 million buildings in the United States. More than 100 insurance carriers use it, and over 300 approved rating and underwriting filings reference CAPE data across 40 states.
CAPE's actuarial findings justify the adoption: roofs rated in severe or poor condition show 2.5 times the wind and hail claims frequency of excellent-rated roofs. Homes where trees overhang the roof experience 90 percent higher wind-related losses. Those numbers come from retrospective analysis across millions of aggregate years of carrier loss data, and they are why state insurance departments have approved CAPE attributes for use in pricing, not just underwriting. Insurers in at least 15 states can legally charge you more because of what an AI saw from above.
Eagleview operates at comparable scale, with a library exceeding three billion images covering 94 percent of the U.S. population. A June 2025 independent benchmark by CompassData confirmed 98.77 percent accuracy for Eagleview's roof measurements, tested via drone LiDAR and terrestrial LiDAR in Denver. In February 2025, Eagleview integrated its Assess drone platform with Verisk's Xactimate, the claims management system used by 80 percent of the top property insurance carriers. Twenty-four of the top 25 U.S. insurers are Eagleview customers.
When your insurer evaluates whether to renew your policy, there is a meaningful chance that an AI has already scored your roof before any human reads your file.
What Goes Wrong
Accuracy for measurements is not the same as accuracy for condition. Eagleview can measure a roof facet to within 1.23 percent of ground truth. But when CAPE or a competitor rates the condition of your shingles from a satellite photo, the model is interpreting color variation, texture patterns, and geometric anomalies to infer structural degradation. Shadows cast by dormers, HVAC equipment, or mature trees can register as dark patches the model reads as damage. Moss, lichen, and algae staining are cosmetically concerning but structurally irrelevant on most roofing materials. Architectural shingle designs with intentional color variance can confuse models trained predominantly on uniform three-tab installations.
California's Insurance Commissioner Ricardo Lara confirmed the failure mode directly: the California Department of Insurance "investigated numerous complaints where flawed aerial imagery led to wrongful cancellations or nonrenewals." In several cases, insurers used "imprecise drone or satellite photos to assess roof conditions, resulting in policies being incorrectly dropped due to erroneous data."
Pennsylvania's Insurance Department issued a bulletin in 2026 noting that some aerial images used for adverse underwriting decisions "barely identify the structure of the home, much less the detailed condition of the roof." Connecticut's insurance regulator advised carriers that aerial images showing discoloration, streaking, or other cosmetic issues alone should not support underwriting action. A Nashville homeowner named William Wilkin got his nonrenewal reversed after his agent demonstrated that tree shadows in the satellite photo made his relatively new roof appear damaged; a roofer's written inspection confirmed the roof was sound.
Five States Are Pushing Back
California Assembly Bill 75, authored by Assemblymember Lisa Calderon and supported by Commissioner Lara, would require insurers to notify homeowners at least 30 days before taking or obtaining aerial images of their property. It prohibits adverse underwriting decisions based on images older than 45 days unless the condition has been verified by an in-person physical inspection conducted within that same window. Homeowners would gain the right to request copies of any aerial images their insurer holds. AB 75 passed the Assembly Insurance Committee 15-0-2.
A companion bill, AB 1559, extends the image-age threshold to 180 days and strengthens the right to dispute AI-generated findings before coverage decisions take effect. Massachusetts and Connecticut issued formal guidance on insurer obligations around aerial imagery in 2025. Pennsylvania's 2026 bulletin stopped short of banning the practice but stated it would be "prudent" for insurers to validate damage with a physical inspection "in the absence of unequivocal and material damage."
At the national level, the NAIC's Model Bulletin on the Use of Artificial Intelligence Systems by Insurers has been adopted by more than half the states. It requires governance programs for AI systems, bias monitoring, data quality validation, and vendor oversight. But it does not specifically address aerial imagery, and it does not mandate homeowner notification.
Why This Is a Building Problem
If you are building a home in 2026, your roof will be photographed from above before your first insurance renewal. Your builder probably does not know this, and your architect almost certainly has never seen the scoring criteria. Roof material, color, geometry, tree proximity, and yard maintenance are all variables the AI evaluates, and none of them appear in the IRC or in any building code your designer references during construction documents.
Consider what this means at the drafting table: a complex roof with multiple dormers, valleys, and varying ridge heights creates shadow patterns that aerial AI may flag as inconsistencies. Standing seam metal in light colors presents a clean, high-contrast surface that scores well. Dark asphalt architectural shingles with intentional color blending can register as granule loss or wear, particularly at low satellite resolution. Landscape architects who plant shade trees within ten feet of the roofline are improving energy performance and potentially degrading the home's aerial insurability score simultaneously.
For buyers, the risk is subtler. You close on a home, insure it, and two years later receive a nonrenewal notice based on aerial images you have never seen. A nonrenewal for roof condition follows you; it affects your eligibility with other carriers, which means the AI's opinion of your roof can cascade into a coverage gap that raises your borrowing costs. Lenders require homeowner's insurance as a condition of the mortgage. If you cannot get coverage, you cannot keep the loan.
What You Can Do Right Now
If you receive a nonrenewal or cancellation notice citing roof condition, request the aerial images immediately. Every insurer contacted for this article acknowledged that policyholders can request the evidence. Compare the photos to current conditions; if the images are outdated or misinterpret shadows as damage, submit a written appeal with a licensed roofer's inspection report and current, close-up photographs. Wilkin's case in Nashville proves this works.
If you are designing a new home, ask your insurer which aerial data vendor it uses and whether it has filed CAPE or equivalent attributes with your state's department of insurance. Request a pre-insurance review of your plans if the carrier offers one. Minimize tree canopy directly over the roofline. Choose roofing materials with uniform surface texture and light reflectance. And document your roof at completion with dated, high-resolution photographs from multiple angles, including aerial drone shots if you can manage it, so you have a timestamped baseline that predates any insurer flyover.
Limitations and Counterargument
Aerial AI has legitimate actuarial value. CAPE's 2.5x claims-frequency differential between poor and excellent roofs is drawn from real loss data across real portfolios. Identifying deteriorating roofs before they fail is good underwriting. Insurers argue, with evidence, that aerial inspection is faster, cheaper, and more consistent than sending adjusters into the field, and that it helps them respond more quickly to climate-driven risks like wildfire and severe storms that have made in-person inspections more dangerous and more expensive.
No published study quantifies the false-positive rate of aerial AI roof scoring across residential properties nationally. CAPE reports confidence scores and reason codes with each rating, but the accuracy of condition assessment, as opposed to measurement, has not been independently validated by a third party in a peer-reviewed study. Until it has, every policy decision based on this technology carries an uncertainty that the homeowner bears and the insurer controls.
What California, Pennsylvania, and Connecticut are doing is not banning the technology. They are requiring that the evidence be shown to the person it affects, that the evidence be current, and that the person get a chance to challenge it. That is not an unreasonable standard for a system that can determine whether you keep your home insurance.