Insurance underwriting is fundamentally a process of discrimination in its purest, statistical sense: the act of distinguishing between different risks to assign appropriate premiums. Even so, the legal and ethical boundaries of how insurers discriminate are strictly defined. The short answer is that insurance company underwriters are allowed to discriminate based on risk factors that correlate statistically with the likelihood of a loss, but they are strictly prohibited from discriminating based on protected characteristics defined by federal and state anti-discrimination laws. Understanding where the line is drawn between actuarial fairness and illegal bias is essential for consumers, agents, and industry professionals alike That's the part that actually makes a difference..
The Core Concept: Risk Classification vs. Unfair Discrimination
To understand the regulatory landscape, one must first distinguish between two types of discrimination The details matter here..
Risk Classification (Actuarial Fairness) This is the bread and butter of underwriting. Insurers group applicants with similar risk profiles—age, driving record, health history, property location, credit-based insurance scores—to predict future claims accurately. If a 16-year-old male driver pays more for auto insurance than a 50-year-old female driver, it is because actuarial data supports a higher probability of accidents for the former group. This is legal, necessary, and considered "fair discrimination" because it aligns price with expected cost Turns out it matters..
Unfair Discrimination (Illegal Bias) Unfair discrimination occurs when an insurer treats applicants with similar risk profiles differently based on characteristics unrelated to the expected loss. As an example, charging two identical drivers different premiums solely because of their race, religion, or national origin constitutes unfair discrimination. Most state insurance codes explicitly forbid rates that are "excessive, inadequate, or unfairly discriminatory."
Federal Legal Framework: The Hard Lines
While insurance is primarily regulated at the state level (under the McCarran-Ferguson Act), several federal statutes create absolute floors that no state law can undercut.
The Civil Rights Act and Fair Housing Act
Title VII of the Civil Rights Act of 1964 prohibits employment discrimination, but it also impacts insurance when coverage is tied to employment benefits. More broadly, the Fair Housing Act (FHA) prohibits discrimination in the sale, rental, and financing of dwellings. This extends to homeowners insurance. An underwriter cannot deny coverage, charge higher premiums, or offer different terms for a property based on the race, color, religion, sex, national origin, familial status, or disability of the owner or the neighborhood's demographic composition. "Redlining"—the practice of denying services to residents of certain areas based on racial demographics—is a direct violation of the FHA.
The Affordable Care Act (ACA) and Health Insurance
The ACA fundamentally reshaped health insurance underwriting. Prior to 2014, individual market underwriters routinely discriminated based on pre-existing conditions, gender, and health status. Today, the ACA mandates guaranteed issue and community rating (with limited adjustments for age, geography, tobacco use, and family size). Health insurers can no longer deny coverage or charge higher premiums because an applicant has diabetes, cancer, or is pregnant. This is the most aggressive federal restriction on underwriting discretion in the industry Nothing fancy..
The Genetic Information Nondiscrimination Act (GINA)
Enacted in 2008, GINA prohibits health insurers and employers from using genetic information to make underwriting decisions. An underwriter cannot request, require, or purchase genetic testing results, nor can they adjust premiums based on a family medical history that suggests a genetic predisposition Less friction, more output..
The Americans with Disabilities Act (ADA)
The ADA prohibits discrimination based on disability. While insurers can still underwrite based on actuarial risks associated with a disability (e.g., a specific condition that statistically shortens life expectancy), they cannot use a disability as a proxy for general uninsurability if the actual risk data does not support it. The coverage must be based on sound actuarial principles, not stereotypes.
State Regulation: The Primary Battleground
Because the McCarran-Ferguson Act grants states the primary authority to regulate the "business of insurance," the specific list of prohibited rating factors varies significantly by jurisdiction. State insurance commissioners enforce statutes that generally mirror federal protected classes but often go further That's the part that actually makes a difference..
Commonly Prohibited Factors Across States
- Race, Color, National Origin, Religion: Universally banned.
- Sex/Gender: This is a complex and evolving area.
- Auto Insurance: Many states (including California, Hawaii, Massachusetts, Michigan, North Carolina, and Pennsylvania) have banned the use of gender as a rating factor. Other states allow it if actuarially justified.
- Life/Health: The ACA banned gender rating in health insurance. In life insurance, gender-based mortality tables are still widely permitted because the statistical difference in life expectancy is profound and consistent.
- Sexual Orientation and Gender Identity: An increasing number of states (e.g., New York, California, Colorado, Washington) explicitly prohibit using sexual orientation or gender identity in underwriting and rating for all lines of insurance.
- Marital Status: Often prohibited in auto and homeowners insurance, though sometimes permitted in life insurance.
- Credit-Based Insurance Scores: This is currently the most contentious rating factor. While the industry argues credit scores are highly predictive of claims frequency, critics argue they proxy for race and income. States like California, Hawaii, Massachusetts, and Michigan have banned or severely restricted the use of credit history in personal lines underwriting. Other states allow it but require transparency and adverse action notices.
The "Disparate Impact" Doctrine: A Legal Gray Area
Even if an underwriting guideline is "facially neutral" (it doesn't mention race), it can still be challenged under the disparate impact theory. Even so, under this legal standard, a policy or practice that has a disproportionately adverse effect on a protected class is illegal unless the insurer can prove:
- The practice is justified by business necessity (a manifest relationship to the insurance risk).
- There is no less discriminatory alternative available to achieve the same business goal.
Quick note before moving on Most people skip this — try not to..
As an example, if an insurer uses a specific ZIP code rating territory that results in significantly higher premiums for predominantly minority neighborhoods, they must prove the territorial boundaries are drawn strictly based on loss costs (crime rates, fire protection, traffic density) and not as a proxy for race. If a broader territory or a different variable (like miles driven) predicts risk just as well without the disparate impact, the insurer may be forced to change its model.
Specialized Lines: Where Discrimination is Explicitly Permitted (or Required)
There are notable exceptions where the law explicitly allows—or even requires—differentiation that would be illegal in other contexts.
Life Insurance and Age
Age discrimination laws generally do not apply to life insurance underwriting. Mortality is inextricably linked to age. Charging a 70-year-old significantly more than a 30-year-old for a term policy is not only allowed; it is an actuarial necessity. The Age Discrimination Act of 1975 explicitly exempts insurance practices based on actuarial data.
Disability and Long-Term Care Insurance
Underwriters can decline coverage or rate up applicants based on specific medical impairments that correlate with higher morbidity or mortality. This is not considered illegal disability discrimination because the classification is based on the medical risk itself, not the protected status of "having a disability" broadly. On the flip side, they cannot deny coverage to a person with a disability unrelated to the risk being insured (e.g., denying a deaf applicant life insurance based solely on deafness, which does not impact mortality) Still holds up..
Annuities and Gender
Because women statistically live longer than men, gender-distinct mortality tables are standard in annuity
pricing. Insurers use gender-distinct mortality tables to calculate payouts: because women have longer life expectancies, they often receive lower monthly payments for the same premium (or pay higher premiums for the same guaranteed income) than men of the same age. While the Affordable Care Act banned gender rating in health insurance, and the EU banned it broadly in 2012, U.S. life and annuity carriers generally retain the right to use gender as a rating variable, supported by actuarial data demonstrating a clear correlation with longevity risk.
Health Insurance: The ACA Framework
The Affordable Care Act (ACA) fundamentally rewrote the rules for health underwriting. It implemented guaranteed issue and community rating requirements, prohibiting insurers from denying coverage or charging higher premiums based on health status, medical history, or genetic information. Premiums can vary only by age (capped at a 3:1 ratio for adults), geography, tobacco use, and family size. This represents a deliberate policy choice to prioritize social equity and access over strict actuarial equivalence, effectively socializing the cost of high-risk individuals across the risk pool Easy to understand, harder to ignore. And it works..
The Algorithmic Frontier: AI, Big Data, and "Proxy Discrimination"
The most contentious battles in insurance discrimination today are not fought over explicit rating variables, but over proxy discrimination embedded in complex algorithms and "Big Data" sources Which is the point..
Modern predictive models ingest thousands of variables—consumer purchasing behavior, social media activity, internet of things (IoT) device data, satellite imagery of properties, and credit-based insurance scores. While any single variable may appear facially neutral, the combination can act as a high-fidelity proxy for protected classes.
- The "Black Box" Problem: Machine learning models (e.g., gradient boosting, neural networks) often identify non-linear interactions that human actuaries cannot easily explain. A model might learn that a specific combination of shopping habits, device type, and commute pattern predicts loss experience with high accuracy—but that same combination correlates almost perfectly with race or ethnicity.
- Regulatory Scrutiny: Regulators are increasingly demanding explainability (XAI) and disparate impact testing before deployment.
- Colorado SB21-169 (effective 2023) is the landmark legislation here. It prohibits insurers from using any external consumer data or algorithm that unfairly discriminates based on race, sex, sexual orientation, or other protected classes. It places the burden on the insurer to prove non-discrimination through quantitative testing (e.g., demographic parity, equalized odds) and governance frameworks, not just actuarial justification.
- The NAIC Model Bulletin on AI (adopted Dec 2023) establishes expectations for governance, risk management, and testing protocols, signaling that "the algorithm did it" is not a valid defense for unfair discrimination.
The Core Tension: Actuarial Fairness vs. Social Fairness
The industry operates on actuarial fairness—the principle that premiums should reflect the expected cost of risk for each individual or homogeneous class. Society, via legislation and regulation, often demands social fairness—the principle that insurance should not reinforce historical structural inequities or deny access to essential financial security based on immutable characteristics Most people skip this — try not to. No workaround needed..
This tension is irreducible.
- Risk Pooling vs. Risk Segmentation: Insurance requires pooling to function, but modern analytics enable hyper-segmentation (a "segment of one"). In real terms, perfect segmentation destroys the cross-subsidy that makes insurance affordable for high-risk individuals. * The "Essentiality" Factor: The regulatory tolerance for segmentation correlates with the "essentialness" of the product. We tolerate almost no risk classification in health insurance (essential for life) and workers' comp (statutory mandate), moderate classification in auto (mandatory in most states), and wide latitude in life, annuity, and commercial lines (voluntary markets).
Conclusion
The law of insurance discrimination is not a static list of banned variables; it is a dynamic boundary line drawn between valid risk differentiation and unfair discrimination. As underwriting migrates from static rate manuals to dynamic, AI-driven decision engines, the definition of "actuarial soundness" is expanding to include a mandate for algorithmic accountability.
Insurers can no longer rely solely on retrospective loss ratios to justify a variable. But they must now demonstrate that their models are dependable against proxy discrimination, that less discriminatory alternatives have been rigorously tested and rejected, and that their data governance prevents the encoding of historical bias into future pricing. The future of underwriting belongs not to those with the most data, but to those who can prove their models distinguish between risk and identity with mathematical precision and ethical rigor.
