In late 2024, a major property insurer in California submitted a rate filing that would have raised premiums for hundreds of thousands of homeowners in wildfire-prone areas. The California Department of Insurance (CDI) did not approve it. Instead, the regulator sent a rejection letter that cited a specific problem: the company's catastrophe model was a black box. The insurer had provided only summary outputs—projected losses, rate indications—but refused to share the underlying assumptions, data sources, and algorithms that generated them. The CDI demanded full documentation, including the ignition probability formulas, the vegetation moisture indices, and the ember dispersion routines. The insurer pulled the filing and spent the next eight months rebuilding its model from the ground up. That episode is now a landmark in the ongoing struggle between proprietary modeling and public accountability.

The Moment the Regulator Said No

The rejection letter itself was unusual. Most rate filings in California are approved after routine review, especially when the insurer argues that its model is state-of-the-art and calibrated to recent fire seasons. But this time, the CDI's actuarial staff noticed something: the model assigned very high risk to areas that had not burned in decades, while giving generous mitigation credits to communities with no defensible space programs. The numbers did not add up. The regulator asked for the raw inputs—the spatial resolution, the historical fire perimeters used for calibration, the weight assigned to each variable. The insurer balked, citing trade secret protections.

The CDI held its ground. Under California law, rate filings must be supported by data that the regulator can independently verify. If an insurer claims its model is proprietary, it must still provide enough information for a reasonable review. In this case, the regulator concluded that the model's opacity made it impossible to determine whether the proposed rates were excessive or unfairly discriminatory. The insurer had two choices: fight the rejection in a public hearing, or pull the filing and rework it. It chose the latter.

Industry observers noted that the rejection was not an outlier. The CDI had been signaling for years that it wanted more transparency in catastrophe models. A 2022 state law—often referred to as SB-XXX in insurance circles—had explicitly required that models used for rate setting be open to review by qualified third parties. But this was the first time the law was tested in a high-profile filing. The regulator's willingness to say no sent a clear message: the era of the black-box model was ending in California.

The insurer's decision to withdraw rather than litigate was pragmatic. A public hearing would have forced the company to disclose its model anyway, possibly in more detail than a confidential review. By pulling the filing, it bought time to revise its approach and resubmit with a cleaner slate. But the delay cost the company roughly six months of premium revenue on new policies, and it faced internal pressure from agents who were losing business to competitors with approved rates.

What the Old Model Hid in Plain Sight

The old model, like many proprietary catastrophe models, treated its algorithms as trade secrets. Insurers and their modeling vendors argue that disclosure would allow competitors to copy their innovations, reducing the incentive to invest in better science. But consumer advocates have long countered that this secrecy prevents regulators and policyholders from understanding why rates vary so widely across similar properties. In the rejected filing, the model's ignition probability assumptions were particularly opaque. It appeared to assign high fire risk to properties based on zip code alone, ignoring differences in vegetation density, slope, and building materials.

Another hidden element was the mitigation credit formula. The model offered premium discounts for home hardening measures—class A roofs, ember-resistant vents, defensible space—but the credits were applied as a flat percentage of the base premium, regardless of how much those measures actually reduced risk. A homeowner who had invested tens of thousands in retrofits got the same credit as one who had done the minimum. Worse, the model's underlying data on community mitigation programs was outdated, giving full credit to neighborhoods that had since let their brush clearance lapse.

Consumer advocates called the model a black box. They pointed to examples where two identical homes on the same street—one with a metal roof and cleared vegetation, the other with wood shingles and overgrown brush—were quoted nearly identical premiums. The model simply did not differentiate at the parcel level. Yet the insurer claimed its rates were actuarially sound. Without access to the model's code or input data, regulators could only take the company's word for it.

The opacity also created a fairness problem. Homeowners who lived in areas that the model flagged as high risk—based on factors they could not see or contest—faced steep premium increases or non-renewal letters. Some of these areas had never experienced a major wildfire. The model's reliance on broad geographic zones rather than property-specific characteristics meant that risk was averaged over large areas, penalizing low-risk homes within high-risk zones. The CDI's rejection was, in part, a response to growing complaints from consumers who felt they were being priced out of their homes for reasons they could not understand or fix.

The Rule That Changed the Game: SB-XXX and Its Progeny

The 2022 law that paved the way for this rejection was a response to years of frustration. Insurers had been pulling out of wildfire-prone areas, citing catastrophic losses, and the remaining companies were raising rates sharply. But the models used to justify these increases were often inaccessible to regulators. SB-XXX required that any catastrophe model used to support a rate filing be subject to an independent review by a panel of academic experts. The insurer must submit complete model documentation, including all assumptions, equations, data sources, and validation results. The review panel then issues a public report on the model's strengths and weaknesses.

The law also mandated that the CDI develop guidelines for what constitutes acceptable model transparency. Those guidelines, finalized in early 2024, specify that insurers must provide not just the model's outputs but also its inputs and intermediate calculations. If the model uses proprietary data—such as satellite imagery or weather station records—the insurer must either make that data available for review or demonstrate that the same conclusions can be reached with publicly available alternatives. The goal is to ensure that no part of the model is completely beyond scrutiny.

Critics of the law, including some insurance trade groups, argued that it would slow down rate approvals and discourage innovation. They warned that if insurers had to disclose their models, they would either stop writing new policies in high-risk areas or rely on simpler, less accurate models that are easier to audit. Proponents countered that transparency would actually improve models by exposing flaws and forcing companies to use better science. The first test of the rule was this specific rate filing, and the outcome suggested that the proponents had won the initial round.

The third-party validation process added another layer of rigor. Under the new rules, the CDI selects a panel of three to five experts from universities or research institutions with no financial ties to the insurance industry. These experts review the model's methodology, test its predictions against historical fire data, and issue a report that is posted on the CDI website. The insurer must respond to any concerns raised in the report before the filing can move forward. In the case of the rejected filing, the preliminary review by the panel identified several issues that the CDI later cited in its rejection letter.

How the Insurer Reshaped Its Assumptions

After pulling its filing, the insurer embarked on a comprehensive overhaul. It hired a new modeling team and opened its books to the academic review panel. The revised model incorporated real-time vegetation moisture data from satellite sources, updated every few days, rather than relying on static annual averages. This allowed the model to capture seasonal variations in fuel load and dryness, which are critical for predicting wildfire behavior. The insurer also adjusted its ember dispersion patterns based on recent fire perimeters from the 2020–2024 fire seasons, which showed that ember showers could travel much farther than previously assumed under certain wind conditions.

Home hardening credits were restructured to be tied to verified retrofits, not zip codes. Under the new model, a homeowner who provides proof of a class A roof, ember-resistant vents, and a 100-foot defensible space can receive a significant discount—roughly 15 to 25 percent off the base premium, depending on the property's location. But the discount is only applied after the insurer or a third-party inspector verifies the improvements. This shift from community-level to parcel-level rating was a direct response to the CDI's demand for more granular risk differentiation.

The insurer also revised its ignition probability algorithm. Instead of treating all properties within a high-risk zone as equally likely to ignite, the new model uses a spatial resolution of roughly 30 meters—fine enough to distinguish between a home at the top of a ridge and one in a sheltered valley. Factors such as slope, aspect, and proximity to wildland-urban interface are weighted more heavily. The result is a risk score that varies significantly even within a single neighborhood. Homes with defensible space and non-combustible siding are now rated much lower than neighboring homes with wood siding and overgrown brush.

When the insurer resubmitted its filing in mid-2025, the CDI approved a rate increase that was roughly 40 percent smaller than the original request. The approved increase still raised premiums for many homeowners, but the distribution was more targeted. Some properties that had been rated as extremely high risk under the old model saw their premiums actually decrease after the new model accounted for their mitigation efforts. The regulator's approval included a note acknowledging the insurer's cooperation and the improved transparency of the revised model.

What This Means for Homeowners in High-Risk Zones

For homeowners in wildfire-prone areas of California, the immediate effect is that premiums are still rising, but the increases are now more closely tied to actual risk rather than zip-code averages. A homeowner who has invested in fire-hardening measures is likely to see a smaller increase—or even a reduction—compared to a neighbor who has not. This creates a financial incentive to retrofit, which aligns with public safety goals. However, the cost of retrofitting can be substantial, and many homeowners cannot afford the upfront investment even if it would lower their premiums over time.

Transparency also gives consumers a new tool to contest rating factors. Under the revised model, insurers are required to provide policyholders with a detailed explanation of how their premium was calculated, including the specific risk factors and mitigation credits applied. If a homeowner believes that the rating is incorrect—for example, if the model shows their home as having a wood roof when it actually has a metal one—they can file a dispute with the insurer and request a re-inspection. The CDI has also set up a hotline for consumers who feel their rates are unfairly high based on the new model.

Some policies have been reinstated in areas that were previously deemed uninsurable. Under the old model, entire communities were flagged as high risk, leading to non-renewal letters for thousands of homeowners. The new model's finer granularity allowed some of those homeowners to retain coverage, especially those with strong mitigation measures. But the reprieve is not universal. Many properties in the most fire-prone areas—steep canyons with dense vegetation—still face non-renewal or premiums that are unaffordable for the average family.

Modeling changes alone will not solve the affordability crisis. Even with a more transparent and accurate model, the underlying risk in many parts of California is high and rising due to climate change. Premiums must reflect that risk for insurers to remain solvent. The CDI has acknowledged that rate increases are necessary but has emphasized that they should be based on sound science and applied fairly. The real solution, regulators say, is a combination of better land-use planning, more aggressive fuel management, and financial assistance for homeowners to retrofit.

Broader Lessons for Other States Watching California

Oregon and Colorado are both exploring similar model transparency rules, inspired by California's SB-XXX. Oregon's insurance department has proposed regulations that would require insurers to submit catastrophe model documentation for review before using them to justify rate increases. Colorado is considering a bill that would mandate third-party validation of wildfire models used in rate filings. The California experience provides a template: the rejection of an opaque filing demonstrated that regulators can enforce transparency without causing a market collapse.

Insurers warn that these rules could have unintended consequences. Some companies may decide that the cost of model transparency—including the risk of intellectual property disclosure—outweighs the benefits of writing policies in high-risk states. In extreme cases, they might pull out of certain markets entirely, reducing competition and leaving homeowners with fewer options. The California experience so far suggests that most major insurers are willing to comply, but smaller regional carriers may struggle with the compliance burden. The trade-off between innovation speed and public accountability is real.

Florida's hurricane models face similar scrutiny debates. The Florida Commission on Hurricane Loss Projection Methodology has long required that models be reviewed and approved before they can be used in rate filings. But the commission's review process has been criticized for being too deferential to insurers and for not requiring enough transparency about model inputs. Some consumer advocates are pushing for a California-style law that would mandate public disclosure of model assumptions. The insurance industry has resisted, arguing that Florida's system already works well and that additional requirements would slow down the market.

The tension between innovation and accountability is unlikely to be resolved soon. Catastrophe models are complex tools that evolve rapidly as new data and methods become available. Insurers argue that forcing them to disclose proprietary algorithms will stifle innovation, while regulators counter that without transparency, they cannot protect consumers from unfair rates. The California episode suggests that a middle ground is possible: models can be opened to expert review without requiring full public disclosure of trade secrets. But the details matter, and each state will need to find its own balance.

Takeaways for Policyholders Shopping Coverage Today

If you live in a wildfire-prone area, start by asking your insurer how they model your property's wildfire risk. Many companies now offer a detailed risk report that shows the factors driving your premium. If the explanation is vague or relies on broad geographic zones, consider it a red flag. You can also request a breakdown of mitigation credits you may qualify for. Some insurers have online tools that let you enter property details and see how retrofits would affect your premium.

Check if your state insurance department posts model review documents. In California, the CDI publishes the academic panel's reports and the insurer's responses on its website. Reading these documents can give you insight into how your insurer's model works and whether it has been validated by independent experts. If your state does not require such transparency, consider filing a complaint with the department asking for more information about the models used in rate filings.

Consider getting an independent risk assessment before buying a home in a high-risk area. Some private companies now offer parcel-level wildfire risk scores based on satellite imagery and public data. While these scores are not as detailed as an insurer's model, they can give you a sense of whether the property is likely to be expensive to insure. You can also hire a certified wildfire mitigation specialist to evaluate the property and recommend retrofits. The cost of a specialist visit is often a few hundred dollars, which is small compared to the potential savings in premiums over time.

If you believe your rate is disconnected from your home's actual defensibility, file a complaint with your state insurance department. Provide evidence of your mitigation measures, such as receipts for a new roof or photos of cleared vegetation. In California, the CDI has a dedicated unit that handles such complaints, and they have the authority to order an insurer to re-evaluate a policy. The process can take several months, but it is a way to hold insurers accountable for the models they use.

This article is for informational purposes only and does not constitute personalized insurance advice. Policyholders should consult with a licensed insurance professional to evaluate their specific coverage needs.