#365: How To Insure Autonomous Vehicles w/Steve Miller of Hub International

“AVs” stands for autonomous vehicles. In this context, it refers to self-driving vehicles that operate with automation rather than a human driver controlling every maneuver.

“DMV” is the Department of Motor Vehicles, the California agency that regulates vehicle licensing and certain testing requirements. The host is describing how the DMV required specific insurance coverage for companies testing autonomous vehicles.

“Every mode of autonomy” is the idea that autonomous systems aren’t limited to cars; they can operate across multiple environments and platforms. The host lists examples like vehicles on wheels, autonomous operation on sea, and in-air autonomy, which matters because insurance risk profiles differ by use case.

“Bias towards founders” here means the tendency to over-trust the people building the technology and assume that novelty automatically reduces risk. The host contrasts that with the reality that insurers still need evidence, data, and risk controls.

“Easy button” is a metaphor for expecting a simple, one-step solution to complex risk and insurance problems. In this segment, the host implies there isn’t a quick fix for insuring autonomous vehicles when data and experience are limited.

In this context, “early adopters” are the first customers willing to buy or insure a new technology before it’s widely proven. For autonomous vehicles, insurers and other risk-transfer partners need enough evidence that the system is safe enough to price the risk responsibly.

This describes a supervised autonomy test regime: even if the vehicle can drive itself, a human driver and/or safety engineer monitors it and can intervene. Insurers care because supervised operation changes the risk profile compared with fully driverless operation.

Drive AI is an autonomous-vehicle company referenced here as an example of an early AV effort. The host mentions writing about it and then notes a later acquisition/hiring by Apple, using it to set context for how supervised autonomy and data-driven assurance work.

Apple is referenced as a company that hired someone connected to Drive AI. In the AV insurance conversation, the mention is mainly to illustrate how major tech firms have invested in autonomous-driving talent and efforts.

“Crash data” refers to recorded information about vehicle collisions and their outcomes. For autonomous-vehicle insurance, crash data is used to build and validate models that estimate how often and how severely incidents occur under different driving conditions.

“Driverless” refers to autonomous operation without a human driver actively responsible for driving. The insurance challenge is that the liability and risk assumptions shift when you move from supervised autonomy to a system that must handle driving on its own.

This describes a regulatory pricing process: insurers (“carriers”) submit their rate structures to state regulators. It matters for autonomous vehicles because the same technology may be priced differently depending on state rules and how regulators accept the risk evidence.

A surcharge is an added fee on top of a base insurance rate. Here, the speaker describes early autonomous-vehicle pricing being pushed higher by applying surcharges based on the limited data available.

Actuaries are insurance professionals who use statistics and risk models to estimate how likely claims are to happen and how much they should cost. In this segment, they’re using driving/vehicle data to set insurance rates for autonomous vehicles.

The bootstrap phase refers to the early period when an autonomous-driving program is still building the evidence, data, and processes needed to operate and insure it confidently. In practice, insurers and brokers have to start with limited real-world performance history and gradually improve risk understanding.

Qualitative vs quantitative risk assessment is the split between judgment-based evaluation (qualitative) and measurements/statistics (quantitative). For autonomous vehicles, early on you may rely more on qualitative evidence (processes, safety engineering artifacts) until enough operational data exists for quantitative modeling.

Non-deterministic refers to outcomes that can’t be fully predicted from the same inputs, because the system may behave differently run-to-run. In autonomous-driving discussions, it’s used to argue that AI-driven behavior can be harder to guarantee and therefore harder to underwrite.

SOPs (standard operating procedures) are documented, repeatable step-by-step processes used to ensure consistent operations. The transcript contrasts the lack of “aircraft level” SOP maturity in autonomy with more established industries where procedures are tightly controlled.

Haul insurance is coverage for transporting goods (often by aircraft in this context), historically used as a way to insure emerging transportation risks. The speaker uses it as an analogy: early insurance products appeared once the industry had enough operational experience to price risk.

The speaker uses “iterative” to describe how autonomous-vehicle insurance and risk assessment evolve over time. Early-stage coverage isn’t just about final system performance; it also depends on testing volume, safety protocols, and real-world operating context. That means underwriting assumptions get refined as more evidence accumulates.

A safety driver is a human in the vehicle who monitors the autonomous system and can take over if needed. Insurers treat this as a key risk factor because it changes how much control is actually delegated to the autonomous system and how quickly hazards can be mitigated. That affects both underwriting questions and expected claim patterns.

Underwriters are the insurance specialists who evaluate risk and decide whether to offer coverage and on what terms. In autonomous-vehicle insurance, they rely on detailed operational and testing information (like where vehicles run and under what conditions) to estimate the likelihood and cost of claims. The transcript emphasizes transparency and getting the right technical people in the room.

Build America 250 is referenced as a policy framework that affects autonomous-vehicle insurance discussions. The speaker’s point is that it doesn’t specify how liability must work, which matters because liability and negligence drive how risk transfer is structured. Listeners should treat it as a named regulatory/program context rather than a technical vehicle feature.

ADS stands for “autonomous driving system.” It refers to the combined hardware and software that performs the driving task (perception, planning, and control). The transcript discusses ADS crashes and how insurers think about claims—often focusing on what the vehicle did to other road users, not just whether the ADS made a “mistake.”

A loss record is the insurer’s internal history of claims and claim outcomes tied to a policyholder or vehicle. In the segment, the point is that if the other party’s coverage doesn’t fully pay (e.g., due to state minimum limits), the remaining cost can still count against the AV’s insurance history.

The human variable refers to the unpredictability and variability introduced by people—especially safety drivers—within an otherwise automated system. The host frames it as a key challenge for insurers and risk assessment because it’s harder to model than purely automated behavior.

Third party validation means independent verification of a system’s claims—here, evidence that an autonomous driving system is safe and operates correctly. The segment frames it as something insurers weigh when deciding how much risk to underwrite.

Simulation is using computer models to test and validate driving behaviors without running on public roads. In the segment, the host contrasts simulation and training with “lip service,” emphasizing that insurers want credible, detailed operational evidence.

A close course (closed course) is a controlled track or test environment where vehicles can be tested without normal traffic. The host lists it as part of the evidence insurers expect when evaluating how a company proves safe operation.

Engage with the marketplace here means actively and constructively working with insurers and other risk stakeholders rather than avoiding or fighting them. The host’s advice is that insurers respond better to clear, cooperative communication than to combative or evasive behavior.

“Robotoxys” appears to be a mis-transcription of “robotaxis,” which are autonomous vehicles operating as ride-hailing services. In insurance terms, robotaxis change the risk picture because the vehicle is expected to drive without a human in the loop.

“Tesla approach” here is shorthand for an iterative software development model: the system improves over time via updates rather than being treated as a fixed, fully validated product. For insurance, that matters because a single software update can change real-world behavior and therefore the risk profile.

A software update is a change to the autonomous driving system’s code or calibration delivered after the vehicle is already in service. The discussion highlights that a single update can dramatically change the vehicle’s risk profile, which complicates how insurers set coverage and pricing.

Actuarial refers to the field of using statistics and risk models to price insurance and estimate claim likelihoods. The host argues that autonomous-vehicle insurance needs actuarial expertise plus engineering knowledge to understand how autonomy behaves in the real world.

Waymo is a brand/company associated with autonomous driving systems. The transcript uses Waymo as an example of a system that may be “the smartest driver,” but then an update can effectively change the behavior—so the insurer’s risk assessment has to account for software change.

In autonomous-vehicle contexts, NVR usually refers to an in-vehicle recording system (often “non-volatile recording”) that stores sensor/drive data for later review. The speaker is saying it can look bad (e.g., low performance or poor metrics) but that the vehicle is still driving many miles, and the system’s behavior changes after updates.

Loss reserves are the amounts insurers set aside to pay for future claims that have already occurred or are expected to arise. In the AV context, the speaker is saying insurers must estimate how claims will ultimately play out, which can lag behind software changes.

A risk transfer partner is the party (typically an insurer or reinsurer) that takes on financial risk in exchange for premiums. The speaker argues that for AV insurance to work, the insurer must understand the risk well enough to price it and still profit.

The judicial system is the legal process that determines how liability is assigned and how claims are settled or judged. The speaker emphasizes that AV risk isn’t only about whether the technology works—it also depends on how courts and legal processes resolve disputes.

Underwriting is the process insurers use to evaluate risk and decide what to cover and at what price. The segment frames underwriting as something that can go wrong when new entrants aggressively sell a “vision” but don’t properly price or manage the risks they’re taking on.

MGA stands for “managed general agent,” an insurance intermediary that can underwrite and manage certain insurance risks on behalf of an insurer. In the segment, the speaker warns that MGAs can be aggressive and may rely on reinsurance papers, which can create risk if they’re focused on growth over responsibility.

The MG MGA is a classic British sports car made by MG, known for its lightweight, open-top (roadster) design and lively driving feel. It’s often discussed in automotive history because it represents the MGA line’s popularity in the mid-20th century and is a common choice for enthusiasts and restorations. In a podcast, it may come up as a recognizable example of classic sports-car engineering and heritage.

Reinsurance is insurance that insurers buy to protect themselves against large losses. The speaker mentions “a reinsurance paper” to describe how some entrants structure risk transfer, and they suggest that this can be dangerous if the underwriting incentives are misaligned.

Private equity refers to investment firms that buy or fund companies with the goal of improving performance and exiting later. In the segment, the speaker suggests that private equity backing can lead to “grow at any cost” behavior, which may increase underwriting and claims risk.

In this insurance context, “aspirational” means the coverage/discount is based on hoped-for outcomes rather than demonstrated performance. The speaker contrasts it with actuarial pricing, implying the insurer is betting on future safety improvements.

“Liability thresholds” are the legal standards or limits that determine when and how responsibility (and damages) apply under insurance and regulatory frameworks. The host says the referenced act does not set up new liability thresholds, meaning it doesn’t fundamentally rewrite the legal bar for liability—though it still changes how AV software is treated.

Subrogation is an insurance process where an insurer that pays a claim can pursue recovery from another responsible party. The host mentions “product liability subrogation,” implying that if an autonomous system malfunction contributes to a crash, the insurer may seek reimbursement from manufacturers or other involved parties.

Cyber liability is insurance coverage for losses arising from cyber events such as hacking, data breaches, or other digital system failures. For autonomous vehicles, the host is linking cyber risk to the vehicle’s software and connectivity—where a cyber incident could contribute to crashes or damages.

In insurance, the loss ratio is the share of premium dollars that end up being paid out as claims (plus related costs, depending on how it’s calculated). A 116% loss ratio means the insurer is paying out more than it takes in, which drives higher auto rates over time.

“Litigious” describes places where people are more likely to sue, which can increase the size and frequency of insurance claims. That legal environment can make auto insurance more expensive, especially for high-exposure technologies or incidents.

“Nuclear verdicts” is a legal-insurance term for extremely large jury awards in personal injury or liability cases. They matter because even a small number of severe outcomes can overwhelm an insurer’s pricing assumptions and worsen loss ratios.

Autopilot refers to Tesla’s driver-assistance system that can take over certain driving tasks, but it still depends on a human driver’s supervision. That “human in the loop” aspect changes how liability is assigned and how insurers model risk.

“Human in the loop” means the system isn’t fully autonomous; a person is responsible for monitoring and/or taking over when needed. For insurance, that matters because it can shift fault and affect how insurers evaluate the likelihood and severity of claims.

“L4” refers to SAE Level 4 autonomy, meaning the vehicle can drive itself in specific conditions without human intervention. “V space” is being used here as a shorthand for the vehicle/autonomous-vehicle risk landscape that insurers model for that autonomy level.

Tail risk is the risk of rare, extreme outcomes—events that sit in the “tail” end of a probability distribution. In insurance modeling, it matters because a small number of catastrophic-like claims can dominate expected losses.

In insurance, “settle” refers to resolving a claim by paying an agreed amount rather than fighting it in court. The speaker is describing how insurers often prefer settlement to avoid legal costs and unpredictable jury outcomes.

ODD (Operational Design Domain) is the specific set of conditions under which an automated driving system is intended to work—like certain roads, weather, speeds, and geofenced areas. Insurers use ODD to estimate how likely the system is to encounter situations it can’t handle.

“Insurance towers” refers to stacking multiple layers of coverage (often via reinsurance and different policy limits) to reach very high total coverage amounts. The speaker is using it to describe how large autonomous-vehicle liabilities can require multi-layer structures.

Self-insurance means the operator sets aside its own funds to cover losses instead of (or in addition to) buying external insurance. The speaker ties it to balance-sheet capacity and risk tolerance.

A captive arrangement is when a company creates or uses its own insurance entity to insure its risks rather than buying coverage from the open market. The speaker frames it as a way for an AV developer to price risk using its own data and risk profile.

A captive structure is the operational setup for that captive insurance approach, where the AV developer (or its captive insurer) sets premiums and reserves loss funds. Because the company is both the insured and the risk holder, it can participate in underwriting profits.

“ODE’s” appears to refer to autonomous driving deployments/operations outside normal public-road use, where insurers can observe incidents and performance without the same exposure as consumer driving. The exact acronym isn’t expanded in the segment, but it’s clearly being contrasted with “strictly on the road.”

The speaker describes companies “pivoting or expanding” from autonomous vehicle efforts into other autonomy domains (industrial, defense, and other modes). The point is that lessons learned in one autonomy area can transfer into underwriting and risk assessment in others.

“Driver in the loop” means the automated system can handle some driving tasks, but a human driver is still expected to monitor the situation and be ready to take over. In insurance discussions, it affects how much risk is attributed to the vehicle automation versus human supervision.

Adaptive cruise control is a driver-assistance feature that automatically adjusts your speed to maintain a safe following distance from the vehicle ahead. The segment uses it to illustrate how the car may continue driving “as if” it’s safe to do so, even when road geometry (like curves) and driver attention become critical.

In this context, FC is referenced alongside AEB as another safety system that has evidence behind it. While the transcript doesn’t spell out the acronym, it’s commonly used in ADAS discussions to refer to forward-collision-related functionality (often paired with braking or collision warning).

AEB means Automatic Emergency Braking. It’s designed to detect an imminent collision and apply the brakes automatically to reduce impact severity or avoid the crash entirely. The speaker contrasts AEB’s clearer safety evidence with other ADAS functions that haven’t yet shown the same level of proven benefit.

A vigilance task is a sustained attention activity where the driver must monitor the system and the road for rare but critical events. The speaker’s point is that waiting for problems and then intervening may not improve safety, because the driver’s attention can degrade over time.

Following distance is the gap between your vehicle and the one ahead, which ADAS can manage using sensors. The speaker specifically mentions L2 helping manage following distance, tying it to risk reduction and driver workload.

L2 refers to SAE Level 2 driving automation, where the car can control steering and speed under certain conditions, but the human driver must remain engaged and ready to take over. The speaker uses L2 to argue that some assistance (like distance and lane keeping) can help, especially for drivers who are not distracted.

“Keep lane” refers to lane-keeping assistance, which uses cameras and/or sensors to help the vehicle stay centered in its lane. In the segment, it’s grouped with following-distance control as an example of L2 features that can reduce workload when the driver remains attentive.

“Errors of omission” are failures to do something that should have been done—like missing a known safety limitation or not accounting for certain scenarios. In insurance underwriting for AVs, the host contrasts omission-type issues with more direct “we need insurance” situations, implying different risk profiles. The point is that not all problems show up as obvious incidents; some are gaps in coverage or understanding.

The Triumph TR4-A (from 1967) is a classic British sports car known for its simple, analog driving feel and period-correct engineering. The host uses it as an example of a car that’s built for the era’s safety expectations—surviving crashes is not the same as being designed to prevent them.

Cameras are a key sensing method for automated driving and ADAS, used to detect vehicles, lanes, pedestrians, and obstacles. The host notes a tradeoff: while cameras can improve safety, they can also increase repair costs when damage occurs (for example, expensive bumper repairs).

Telemetry is vehicle-generated data (from sensors and control systems) that can be recorded and later used to reconstruct events. Here, telemetry is described as helping determine when a tractor-trailer began braking and how long it took the following driver to react—useful for resolving liability and fraud.

GPU here refers to the graphics processing unit used for heavy parallel computing tasks, such as running perception algorithms for cameras and sensors. The host ties it to the idea that the vehicle can compute relative speeds and other situational facts that support claim resolution.

“Fraud quotient” is an insurance-industry way of describing how much claims are driven by dishonest behavior or disputes. The host argues that better evidence from automated sensing (cameras/telemetry) makes “he said, she said” situations less common, lowering the incentive for fraud.

In insurance, the “float” is the money an insurer collects from premiums before it has to pay out claims. Insurers invest that time gap, and the investment returns can affect profitability and pricing decisions.

A “state insurance fund” is a government-run insurance pool created to ensure coverage when private insurers are unwilling or when risk is too high. The transcript connects it to disaster-prone states, where residents still need a way to obtain insurance.