Does California's AB 311 Secretly Let AI Set Your Auto Insurance Premium

A single missing word, "AI," is sitting at the center of one of the most consequential auto-insurance debates in California history.

Digital Insurance reports that AB 311, the Consumer Driving Data Protection Act, would let insurers use telematics data to set rates, but the bill's author and a prominent consumer advocacy group fundamentally disagree on whether artificial intelligence can be part of that process. That ambiguity matters for every California driver, because the answer determines how much algorithmic power insurers could hold over your premium. According to the Save Max Quote Index, drawn from 3.3 million+ real quote requests, California consistently ranks among the most expensive and most restricted auto-insurance markets in the country, making any regulatory shift here a high-stakes event for policyholders.

A bill that rewrites California's auto-insurance playbook, with a key word missing

AB 311 was first introduced in January and most recently amended by the State Senate on July 9, according to Digital Insurance. Its formal name, the Consumer Driving Data Protection Act, sounds protective. But its operational heart is permissive: it would open the door for auto insurers to collect telematics data from California drivers and use that data directly in rating decisions.

The controversy is not what the bill says. It is what the bill does not say.

The word "AI" appears nowhere in the legislation. Terry Schanz, chief of staff for state assemblymember Tina McKinnor, the bill's author, argues that omission is intentional and dispositive. Consumer Watchdog, a Los Angeles-based consumer advocacy non-profit, argues the opposite, saying the bill's language is broad enough to let AI in through the back door.

That interpretive gap is what makes the California telematics bill AI question so urgent, and so unresolved.

What AB 311 actually says about scoring models

The bill defines a "scoring model" as "a computational, statistical, actuarial, or algorithmic methodology capable of evaluating telematics data, or the inferences derived from those methodologies, to generate a numerical score or predictive assessment used directly or indirectly in rating automobile insurance."

Read that definition carefully. It covers computational, statistical, actuarial, and algorithmic methodologies. It covers inferences derived from those methodologies. It covers assessments used "directly or indirectly" in rating.

What it does not do is draw a bright line excluding machine learning or large language models.

On the consent side, the bill does include a clear protection: insurers must obtain opt-in consent from drivers before collecting telematics data. That means no passive enrollment. You would have to affirmatively agree before your braking patterns, acceleration habits, or nighttime driving frequency could flow into a scoring model.

For California auto insurance shoppers, opt-in consent is meaningful in theory. The question consumer advocates are raising is whether consent addresses the downstream risk of what happens to that data once collected.

Two readings, one bill: how the author and consumer advocates disagree

Schanz is direct. "There is no AI usage in the legislation or in the application of the law," he said. He frames telematics as a straightforward technology: it "simply records a driver's actions and uses that data to determine a driver's insurance rate." He points to the fact that the technology is already used in 49 other states as evidence that it is a known, bounded tool.

Carmen Balber, executive director of Consumer Watchdog, sees the bill differently. She argues that the scoring model definition, which the group also calls "predictive algorithmic modeling," could include the application of AI to produce predictions about auto insurance risks for setting premiums.

"Once a large language model has been trained on data, deleting the data doesn't remove the data's influence on the LLM," writes Justin Kloczko in a Consumer Watchdog article.

That is the crux of the consumer advocate position: even if raw telematics data is eventually deleted, its influence on a trained model persists. The bill does not address that specific scenario.

The two sides are not debating a technicality. They are debating the entire scope of what insurers can build with California drivers' driving data.

How telematics data could feed future AI models, even after deletion

Consumer Watchdog's concern goes beyond current scoring models. The organization warns that collected telematics data could be used to train scoring models or algorithms developed in the future, not just those in use when the bill passes.

This is the LLM training-data problem applied to insurance. When a large language model, or any machine learning system, trains on a dataset, the statistical patterns from that data become embedded in the model's weights. Deleting the original data files afterward does not erase those learned patterns.

Justin Kloczko, writing for Consumer Watchdog, articulates this plainly:

"Once a large language model has been trained on data, deleting the data doesn't remove the data's influence on the LLM."

AB 311 does not appear to contain provisions that address this forward-looking risk. The bill's opt-in consent requirement covers collection. It does not, based on the available language, govern what a scoring model built on historical telematics data retains after that data is nominally purged.

For drivers in states that have already grappled with telematics regulation, like those reviewing Oregon auto insurance rules or Nevada auto insurance frameworks, this is a familiar tension: consumer rights language written for today's technology can create blind spots for tomorrow's.

California vs. the other 49 states: why segmentation matters for your premium

Katie Klutts Wysor, principal in the insurance practice at PwC, offers the clearest explanation of why AB 311 exists in the first place.

California currently restricts the modeling techniques insurers can use to segment policyholders by risk. The result, Wysor explains, is a flattened market where good drivers subsidize higher-risk drivers because carriers cannot price the difference with precision.

"California is a very hard state to do business in for auto insurers. You can't really segment. Therefore, if you're not segmenting well on pricing, you're at risk of attracting business that you typically in any other state would charge more for, because of the risk profile, because the rate to risk is actually higher," Wysor said. "But there's no way for you to get to that rate to risk in California. There's modeling techniques you're not allowed to use. The result is so few people are really good risks. The whole thing is flattened out because you can't get the right segmentation in the book."

Here is what that looks like in practice:

AB 311 would change the California column. Whether that benefits you depends entirely on where your driving behavior places you on the risk spectrum.

The SMQI consistently reflects that California drivers face a compressed pricing environment, which can feel fair until you realize that low-risk drivers are effectively cross-subsidizing higher-risk policies. Telematics-based segmentation, in theory, corrects that.

Who holds the oversight power if AB 311 passes

Even if AB 311 becomes law, it does not remove regulatory oversight. The California Department of Insurance, the state's insurance regulator, would retain the authority to review how insurers apply predictive algorithms to auto insurance risks and rates.

Wysor confirmed this point explicitly: the CDI's review power survives the bill's passage.

This is a meaningful safeguard. California's Department of Insurance has historically been one of the more active state regulators in scrutinizing insurer pricing methodologies. If a carrier's scoring model produces outcomes that appear discriminatory or that rely on prohibited factors, the CDI retains tools to intervene.

The open question is whether the CDI has the technical capacity and regulatory bandwidth to audit AI-influenced scoring models at scale, particularly if the models themselves were trained on data that has since been deleted. That capability question is separate from the legal authority question, and AB 311 does not appear to resolve it.

Drivers in Washington State and Colorado have watched their own state regulators grapple with algorithmic pricing oversight, and the experience suggests that legal authority to review models and practical ability to audit them are two very different things.

What this means for you

If AB 311 passes, review any telematics opt-in offer from your insurer carefully before agreeing, because your consent is the bill's primary consumer protection mechanism. Ask your insurer directly whether their scoring model uses machine learning or AI components, since the bill's definition does not foreclose those technologies. Check back with the California Department of Insurance for any guidance or rulemaking it issues on predictive algorithm review standards after enactment. If you are a safe driver, telematics-based segmentation could lower your premium relative to today's flattened market, but only if your insurer's scoring model is built and regulated fairly.

FAQ

What is AB 311 and what does it do?

AB 311, the Consumer Driving Data Protection Act, is a California bill first introduced in January and most recently amended by the State Senate on July 9. It would allow auto insurers to use telematics data collected from drivers to set insurance rates, using a "scoring model" as defined in the bill. Insurers must obtain opt-in consent from drivers before collecting that data.

Does AB 311 allow AI to be used in auto insurance pricing?

That is the central dispute. The bill's author, assemblymember Tina McKinnor, through chief of staff Terry Schanz, says no AI usage is permitted. Consumer Watchdog, a Los Angeles-based non-profit led by executive director Carmen Balber, argues the bill's scoring model definition is broad enough to include AI-driven predictive modeling. The bill itself does not use the word "AI."

What is a scoring model under AB 311?

The bill defines a scoring model as "a computational, statistical, actuarial, or algorithmic methodology capable of evaluating telematics data, or the inferences derived from those methodologies, to generate a numerical score or predictive assessment used directly or indirectly in rating automobile insurance." Critics say this definition could encompass machine learning systems.

Will the California Department of Insurance have any oversight role?

Yes. According to Katie Klutts Wysor, principal in the insurance practice at PwC, the California Department of Insurance would retain authority to review how insurers apply predictive algorithms to auto insurance risks and rates even if AB 311 becomes law.

Why does telematics matter more in California than in other states?

California currently restricts the segmentation tools insurers can use to differentiate risk. Wysor explains that this prevents carriers from matching rates to actual risk profiles, which flattens pricing and can result in good drivers subsidizing higher-risk policyholders. The 49 other states already permit telematics-based rating, which allows more precise risk segmentation.

About Kyle Greenwood

Kyle Greenwood is a Writer and Researcher at Save Max Auto with a decade of consumer-content experience. He specializes in explainers, longer-form features, and Q&A guides on the topics auto drivers actually search for. Read more from Kyle Greenwood →

Edited by Cassidy Richey.

Methodology

This article is grounded in the source linked above. Save Max Auto data points referenced here are drawn from the Save Max Quote Index (SMQI), a proprietary instrument reflecting 3,364,317 real consumer quote requests submitted to savemaxauto.com. State and carrier rankings reflect the lifetime dataset; year-over-year shifts reflect a rolling 12-month window. The index is refreshed monthly. External authority figures referenced (NAIC, NHTSA, state regulators) reflect the most recent public data releases available at time of writing.

Sources

• Primary source: Digital Insurance, "Why California's telematics bill leaves insurers guessing on AI"