⚖️ Policy
The AI That Decides Whether You Can Insure Your Home Won't Tell You the Score. One State Just Changed That.
Steve Poizner spent tens of thousands of dollars hardening his home against wildfire. He replaced the roof, cleared defensible space, installed ember-resistant vents. He did exactly what his insurer told him to do. They non-renewed him anyway, without explanation and without offering any mechanism for appeal.
Poizner is a former California insurance commissioner. If even the state's former top insurance regulator can't get a straight answer about why an AI model decided his home was too risky, what chance does a first-time homebuyer in the San Bernardino foothills have?
Right now, in most of the United States, the answer is: almost none, because no state except Washington requires disclosure, no federal regulation mandates transparency, and there is no right to see the algorithm that priced you out of your own home, to understand which of the fourteen satellite-derived variables pushed your score past the threshold where your insurer decided the risk was no longer worth bearing, or to challenge the assessment with evidence that the model failed to incorporate.
The score you never see
Every major home insurer in wildfire-exposed states relies on third-party AI wildfire risk scores. Companies like Verisk, ZestyAI, and Property Guardian use satellite imagery, property-level data, historical fire patterns, and fuel condition modeling to assign a risk score to individual homes. Verisk's FireLine system scores properties from 0 (negligible) to 30 (extreme). These scores determine whether you get a policy, what you pay, and whether your coverage survives renewal season.
Sophistication is not the problem. Property Guardian's Wildfire Recurrence Risk Score incorporates time-aware fire history and evolving fuel dynamics, and during the January 2025 Los Angeles wildfires it identified the areas of highest destruction two to three days before fire even arrived, according to the company, outperforming every static hazard map in the industry. Through partnerships with EigenRisk and Convr announced in June 2026, that intelligence now flows into commercial underwriting dashboards where analysts can drill into property-level risk factors, fuel condition timelines, and destruction probability curves in real time.
Commercial property underwriters can drill into this data in real time, filtering by fuel condition trends, structure-level vulnerability metrics, and time-since-last-burn decay curves that update with every new satellite pass over their portfolio. Homeowners get none of it, just a letter.
$940 billion in blind spots
A January 2026 analysis by ZestyAI found that 1.2 million California properties are rated as high wildfire risk by AI models but show as low risk or carry no rating at all on FEMA maps. Combined property value in those blind spots reaches $940 billion, calculated using a conservative $800,000 median home price, and 300,859 of those homes were built before 1980, predating every modern fire-resistant building code California has enacted.
Only 37.5% of California properties carry any FEMA wildfire rating at all. Put differently: nearly two-thirds of all residential property in the most wildfire-prone state in the country has no federal wildfire assessment whatsoever. What separates the AI models from the federal maps is not a marginal reclassification but a wholesale disagreement about where wildfire risk exists. And the AI is almost certainly right. Count on it.
For homeowners, the practical consequence is this: you can look up your FEMA flood zone in three minutes. You cannot look up the AI wildfire risk score your insurer used to price your policy, or to decide it would rather not write one.
Mitigation the model sees, the insurer ignores
ZestyAI's data contains a quietly devastating finding. Among properties the AI scores as high-hazard (score 7 or higher on its 10-point scale), 41.3% reduced their home-destruction risk through visible mitigation measures between 2022 and 2025. These homeowners cleared brush, upgraded roofing, created defensible space, and the AI tracked every improvement across successive satellite passes.
California regulation does not require insurers to factor any of those mitigation efforts into non-renewal decisions.
Consumer Watchdog, the Los Angeles-based advocacy group, flagged this loophole in written public comments to Insurance Commissioner Ricardo Lara's office three separate times between 2020 and 2022. Commissioner Lara's regulation, submitted to California's Office of Administrative Law in 2022, preserved it intact, which means insurers can require mitigation as a condition of continued coverage, watch the homeowner spend tens of thousands of dollars completing every item on the list, and then non-renew the policy anyway without accounting for whether any of that work was done.
Among extreme-hazard properties, the gap widens further. ZestyAI found that 42.5% of homeowners with scores of 9 or higher improved their property-level risk, but 52% saw the surrounding regional wildfire hazard worsen. Average hazard score increases of 1.3 points translated to a 51% jump in annual wildfire probability. A homeowner can harden the structure but cannot control the surrounding landscape, and while the AI scores both dimensions simultaneously, regulatory frameworks in most states address neither.
Where the displaced go
California's FAIR Plan, the state's insurer of last resort, was not designed to carry this load. In September 2020, it had 202,897 dwelling policies. By September 2024, the count reached 451,799, a 123% increase in four years. Written premium tripled from $424 million in fiscal year 2021 to $1.267 billion in fiscal year 2024, which would rank the FAIR Plan as the state's third-largest home insurer by premium volume. A plan designed as a backstop for the uninsurable fringe of the market is now the third-largest home insurance operation in California, covering properties that private carriers spent decades profitably insuring before their algorithms told them to stop. By September 2025, active policies hit 642,010, of which roughly 580,000, or 90%, were concentrated in counties FEMA rates as "Relatively High" or "Very High" for wildfire risk, according to a MoneyGeek analysis cross-referencing FAIR Plan enrollment with the National Risk Index.
Contra Costa County, not historically a wildfire hot spot, saw its FAIR Plan enrollment increase from 1,094 policies in 2021 to 12,837 in 2025. That is a 1,073% increase in a county most Californians associate with suburban commuter housing, not wildfire zones. AI models appear to be catching risk that old maps never reflected, and insurers are acting on those scores by withdrawing from 46 of California's 58 counties.
FAIR Plan policies provide less coverage than standard homeowners insurance. They exclude theft, liability, and water damage. Homeowners who need those protections must purchase a separate difference-in-conditions policy from a private insurer, if one will write it.
One state forced the question
In February 2026, Washington State's Senate passed SB 5928 on a 48-1 vote. Requested by Insurance Commissioner Patty Kuderer and prime-sponsored by Republican Senator Judy Warnick of Moses Lake, SB 5928 requires insurers that use wildfire risk scores to disclose the score to the homeowner, explain the factors behind it, and provide plain-language steps the homeowner can take to improve it.
It also creates an appeal right. Homeowners who have completed mitigation work since their last evaluation, or who identify demonstrable inaccuracies in their current score, can challenge the assignment directly with the insurer. Insurers must acknowledge any appeal in writing within ten calendar days and respond with a reconsideration decision within thirty.
Washington Realtors, the state's Fire Chiefs Association, and the Confederated Tribes of the Colville Reservation all testified in support. Only one senator voted no.
"If your home is being evaluated and there's financial consequences for you, you deserve to know where that score came from and what you can do to improve it," Commissioner Kuderer said, articulating a principle that forty-eight of her state's forty-nine voting senators apparently found unobjectionable.
What a builder or buyer should do now
If you are building in a wildfire-exposed area, the AI score on your finished home will determine your insurance options more than any other single variable except geography itself. You cannot currently access most commercial wildfire risk scores directly, but you can take steps that the models are designed to detect.
Build with Class A fire-rated roofing, the single highest-impact variable in most property-level wildfire vulnerability models according to Nearmap's Betterview platform. Install ember-resistant attic and soffit vents (mesh no larger than 1/8 inch). Use noncombustible siding within five feet of grade. Clear Zone 0 (0-5 feet) of all vegetation and combustible materials. Maintain Zone 1 (5-30 feet) with fire-resistant landscaping. Document everything with dated photographs, because when the transparency laws eventually arrive in your state and you file an appeal challenging your wildfire risk score, the dated evidence of mitigation work will be the difference between a successful reassessment and a form-letter denial.
If you are buying an existing home, ask your insurer directly for the wildfire risk score used in underwriting your policy. In most states they are not required to provide it, but some will. If your state has adopted or is considering transparency legislation modeled on Washington's SB 5928, your leverage improves. Check California's CDI, Colorado's Division of Insurance, and Oregon's Department of Consumer and Business Services for state-specific rules.
If your policy is non-renewed, request the specific score and scoring factors in writing. California's Prop 103 nominally gives the Department of Insurance the authority to require disclosure. Whether that authority is exercised is a different question.
Limitations
This analysis relies on ZestyAI's proprietary model outputs for the 1.2 million property blind-spot figure. We cannot independently verify the model's classification methodology or validate individual property scores. FAIR Plan enrollment figures are drawn from CDI and Milliman public filings. MoneyGeek's 90% concentration figure depends on a cross-reference with FEMA's National Risk Index, which itself uses county-level rather than property-level risk assessments. Property Guardian's prediction of LA wildfire destruction patterns is based on the company's own post-event analysis, and while the two-to-three-day advance identification is a striking claim, we found no independent academic or regulatory validation of the methodology, the sample size, or the specific accuracy metrics against which the prediction was measured. Washington State's SB 5928 had passed the Senate at time of reporting; its status in the House was not independently confirmed. California CDI regulatory actions referenced date from 2020-2022 public filings; more recent regulatory activity may exist that is not reflected here.
The counterargument
Insurers make a defensible case that property-level AI scoring is a significant improvement over the old system, which assessed risk at the ZIP-code or community level. Under geographic averaging, a fire-hardened home in a high-risk area paid the same rate as a neglected home next door, or both lost coverage entirely. AI scoring should, in theory, reward mitigation at the individual property level. Some insurers are beginning to do so. Nearmap's Betterview platform, for example, distinguishes between wildfire hazard (location-driven, mostly uncontrollable) and wildfire vulnerability (property-driven, largely mitigatable). A homeowner with low vulnerability in a high-hazard area is a fundamentally different risk from a neighbor with high vulnerability in the same location, and better data, deployed transparently, should ultimately produce better outcomes for everyone, including the insurers who gain a more accurate view of their exposure.
None of this is a technology problem. Disclosure requirements, appeal processes, and regulatory accountability simply have not caught up with models that evolved from ZIP-code averages to satellite-tracked, property-level, time-series wildfire intelligence in less than five years. A 48-1 vote in Washington suggests there is not serious opposition to fixing that. What remains unresolved is how many other states will follow Washington's lead, and whether California, where 642,010 homeowners now depend on the FAIR Plan and that number grows every quarter, will close the loophole that lets insurers score a homeowner's improvement and ignore it at the same time.