States Tighten Scrutiny of AI-Based Insurance Pricing Fairness

Why State Regulators Are Rewriting the Rules on Insurance Pricing

State regulators are moving from debate to rulemaking over whether the machine-learning models U.S. insurers use to price coverage are fair. Colorado is the furthest along: Senate Bill 21-169, passed in 2021, requires testing of pricing models in private passenger auto insurance, and the state's regulator is still defining exactly what that testing will involve. New York has issued a circular on model governance and testing that insurers are now studying closely, while Washington, D.C. is developing its own approach and Washington state has a research project under way.

Auto insurance sits at the center of the fairness debate because it is legally required and nearly universal, yet premiums vary sharply from driver to driver. Carriers have adopted advanced modeling, machine learning and artificial intelligence to differentiate risk more precisely — a change that can align prices more closely with expected losses, but also raises questions about disparate impact and proxy discrimination. "Every driver in the U.S. needs auto insurance, but there is great variety in pricing from driver to driver. The inequality in pricing makes the potential for bias and disparate impact a frequent topic of conversation," said Gary Wang, senior consulting actuary at Pinnacle Actuarial Resources.

The regulatory interest is amplified by an affordability squeeze. Premiums have risen faster than the typical cost of living in recent years, driven by higher claim severity and rising labor and parts costs. Wang notes that some budget-constrained policyholders respond by dropping coverage entirely, while most consumers view insurance as a bilateral contract rather than membership in a shared risk pool — a mental model that makes rate increases hard to explain even after a claim-free year.

The analysis was produced by the R&I Brand Studio, the advertising unit of Risk & Insurance, in collaboration with Pinnacle Actuarial Resources, and the publication's editorial staff played no role in its preparation. The same article also announced that Berkshire Hathaway Specialty Insurance promoted Jarek Chmielowski to Healthcare Institutions Portfolio Manager and Greg Struhar to National Underwriting Manager, Healthcare Institutions, both effective immediately, within its healthcare professional liability business.

The Precision vs. Fairness Trade-Off Reshaping U.S. Auto Insurance

Wang's central observation is that the same technology driving regulatory scrutiny also makes the fairness problem harder to read from the outside. Better risk evaluation allows insurers to match prices more closely to expected losses, but the cost of that precision is greater segmentation and differentiation. "When you simply compare premiums side by side, they appear more unequal than ever," he said. Regulators therefore face a genuine trade-off: a model that is fairer in actuarial terms can look more unfair in the distribution of quoted rates.

Colorado and New York Set the Pace — Without a Finish Line

Colorado's SB 21-169 passed in 2021 but has taken years to move into regulation, and the state is still working out what testing will measure — including how premium differences compare with the loss differences observed for the same groups. New York's circular has pulled industry attention because insurers are unsure precisely what governance and testing it demands. With D.C. and Washington state still studying the issues, Wang expects the rules to keep evolving: "Whatever we reach in terms of what it means to govern and test isn't going to be the final answer, but simply a first iteration of an answer." For carriers, that means compliance is an iterative process rather than a one-time fix.

The Affordability Crisis Behind the Complaints

The fairness debate is partly an affordability debate. Rapid premium growth — from higher claim severity and costly labor and parts — pushes consumers to question whether rates reflect their actual records, since most drivers will go years without an at-fault claim. Wang's point that premiums can remain high even in claim-free years highlights the tension between actuarial pricing and consumer perception. If regulators respond by restricting pricing differentiation, insurers could lose the ability to price risk accurately, potentially shifting costs across policyholders in ways that are themselves regressive.

Actuaries as Translators Between Lawmakers and Models

Wang argues actuaries need to be engaged early to define what proxy discrimination actually means, how it should be measured, and how testing can be both practical and accurate. He points to telematics as an example of the nuance: driving data can support behavior modification for some policyholders, while others have limited choice about where they live or when they drive. The constructive question, he says, is how to make those environments safer rather than simply charging more. For consulting firms like Pinnacle, which is already involved in the Washington state study, this regulatory wave represents a growing line of business advising both governments and insurers.

What Insurers, Regulators and Actuaries Should Watch in the Testing Wave

For insurers, regulators and the actuarial profession, the next phase of the fairness debate will be defined by testing methodology, model governance and consumer communication.

• Auto insurers writing in Colorado should treat the SB 21-169 testing regime as an unfinished rulemaking and track the regulator's forthcoming definitions of what testing must cover, including how premium differences are compared with loss differences.
• Carriers using machine-learning rating models should map their model governance and testing documentation against the New York circular's requirements now, rather than waiting for exam cycles to reveal gaps.
• Insurers should quantify the precision-versus-dispersion trade-off Wang describes: as models become more accurate, quoted premiums will diverge more visibly, so companies need a strategy for explaining rate differences to consumers and legislators.
• Regulators should define measurable milestones for testing standards — Wang's "first iteration" framing — so that both industry and the public can see progress as methodology matures.
• Actuaries and consultants have a concrete opening in jurisdictions still studying the issues: D.C. and Washington state have not fixed their approaches, creating demand for practical definitions of proxy discrimination and testing methods.