“AVs” means autonomous vehicles—cars or other vehicles that can drive themselves using sensors and software. Here, the discussion is about insuring those self-driving systems.
“DMV” is the government office that handles vehicle rules and licensing. In this episode, it’s also setting requirements for companies testing self-driving vehicles.
An “R&D application for on-road testing” refers to applying to conduct research and development trials on public roads. For autonomy companies, these approvals often come with regulatory and insurance requirements because real-world testing carries real safety risk.
He’s saying autonomy isn’t just self-driving cars—it can apply to many kinds of vehicles and environments. Insurance has to account for those different scenarios, not just one type of vehicle.
He’s talking about a common mindset: people assume the founders’ new tech is so advanced that it must be safer or easier to insure. The episode argues that insurers still look for proof and risk management, not just confidence in the inventors.
“Liability and risk transfer” is the idea that responsibility for accidents or losses can be shifted—often via insurance or contracts—from one party to another. The host argues that even with new autonomous technology, liability doesn’t automatically disappear or get fully “disrupted” away.
This is an expression meaning “there’s a simple fix.” He’s saying insuring self-driving tech isn’t that straightforward—you can’t just press one button and be done.
Early adopters are the first people or companies willing to try something new. With autonomous vehicles, insurers want proof it’s safe before they commit to covering it.
Term
AV
“AV” stands for autonomous vehicle. Here, the host describes an early testing/validation setup where the vehicle is being tested with a human driver and safety engineer present, and the goal is to show the system is at least as safe as a human in that role.
That’s how early self-driving tests are often done: people are still in the car to watch and take over if something goes wrong. Insurance pricing depends on whether the system is truly driverless or still supervised.
Drive AI is a self-driving company mentioned as background. The point is to connect early self-driving development and how insurers might evaluate safety using real-world data.
Apple is mentioned because it hired someone involved with an autonomous-driving company. It’s part of the “how we got here” context for self-driving development.
“Driverless” means there’s no human actively driving the car. From an insurance standpoint, that’s a big change because the car’s system is responsible for what happens.
Insurance companies have to set their prices according to rules in each state. That means autonomous-vehicle insurance can vary by location because the approval process is state-by-state.
A surcharge is an extra charge added to the insurance price. It’s used to make the cost higher when the insurer thinks the risk is higher or the data is uncertain.
Actuaries are the people at insurance companies who do the math to estimate risk. They use data to figure out how expensive claims are likely to be, so the company can price insurance correctly.
To determine a claim “forensically” means using detailed evidence and analysis—often from vehicle sensors and logs—to reconstruct what happened. The speaker ties this to autonomous vehicles’ onboard sensors and compute, which can support more precise investigations after incidents.
This phrase is describing a hype cycle pattern: early excitement is followed by disappointment (“trough of disillusionment”), then later stabilization and more realistic progress. The speaker uses it to explain how autonomous-vehicle vendors evolve over time as expectations change.
A safety case is a structured argument (with evidence) showing that a system is acceptably safe for its intended use. In autonomous-vehicle insurance, it’s used alongside test results and real-world driving data to help insurers and risk models understand how safety performance translates into claim likelihood.
On-road data refers to real-world driving records collected from autonomous vehicles operating in traffic. The speaker contrasts it with simulation, noting that insurers need both to build credible risk estimates for claims.
Simulated data is information generated by running scenarios in a computer model instead of relying only on real-world driving. For autonomous vehicles, simulation helps cover rare edge cases and produces large datasets that can be combined with on-road evidence for risk assessment.
The bootstrap phase is the early stage where you’re still getting things off the ground. You don’t yet have lots of real-world proof, so it’s harder to judge risk and set up insurance confidently.
This is about using either opinions and expert judgment (qualitative) or numbers and measurements (quantitative). With new autonomous systems, you often start with more judgment until you have enough data to use solid statistics.
Non-deterministic means the same situation might not always produce the exact same result. The concern is that AI behavior can vary, which makes it tougher to predict risk for insurance.
SOPs are written instructions for how to do something the same way every time. The point here is that autonomy is newer, so it may not yet have the same level of standardized procedures as highly regulated industries.
Haul insurance is insurance for moving cargo from one place to another. The speaker is saying that even aviation started by insuring new kinds of risk before everything was perfectly predictable.
Concept
iterative
“Iterative” here means the process improves step-by-step. Instead of judging the self-driving system only by its final results, insurers look at how it’s being tested and how safety is handled as the program develops. Over time, the risk picture becomes clearer.
A safety driver is a person sitting in the car to watch the self-driving system and be ready to intervene. If a human is present and can take over, the risk profile can be different. Insurers ask about this because it affects how likely incidents are and how severe they might be.
Underwriters are the insurance experts who decide how risky a situation is and what the insurance should cost. For self-driving cars, they want lots of details about how the system is tested and used. Their job is to translate that information into coverage terms.
Concept
Build America 250
Build America 250 is a named government program or rule the hosts are discussing. The key takeaway here is that it doesn’t spell out exactly how responsibility (liability) should be handled after a crash. That’s important because insurance depends on who’s legally at fault.
ADS means the car’s self-driving system—the computer and sensors working together to drive. When people talk about ADS crashes, they’re talking about incidents involving that system. Insurers look at the whole situation, including who got hit.
Minimum state limits are the lowest required liability coverage amounts under a state’s auto insurance laws. The host uses California’s minimums (15,000) to show that if the other driver’s coverage caps out, the remaining damage cost can fall back onto the AV’s insurance.
A loss record is basically the insurer’s log of past insurance claims. If a crash isn’t fully covered by the other driver, part of the cost can still show up as a claim on your insurance history.
The human variable means the self-driving system isn’t the only factor—people can behave differently and make different decisions. That makes risk harder for insurers to predict.
Third party validation means someone independent checks and confirms the safety claims. Insurers use that kind of evidence to judge how risky the system is.
Simulation means testing the system in software before real-world driving. The point here is that insurers want more than just saying you did simulations—they want to see how you actually operate safely.
A closed course is a controlled test track where you can run scenarios safely without regular traffic. The host is saying insurers care about whether you’ve tested in realistic, controlled ways.
Concept
engage with the marketplace
This means work with insurers in a straightforward way. The host is saying being difficult or secretive won’t help you long-term.
This sounds like “robotaxis,” meaning self-driving cars that give rides like a taxi service. Because they’re meant to drive themselves, insurers have to think differently about risk.
They mean a “keep improving it with software updates” strategy. If the car’s behavior can change after an update, the insurer has to reassess what could go wrong.
A software update is a change to the car’s computer programs after you’ve bought it. If that update changes how the car drives, the insurance company may need to rethink the risk.
Validation efforts are processes used to prove that an autonomous system performs safely and as intended. In insurance, validation is important because insurers need evidence about how the system behaves across scenarios, not just marketing claims.
Waymo is a self-driving car brand. The discussion says even if a system is great today, an update can change how it drives, which affects insurance risk.
Term
NVR
NVR here means a system that records driving/sensor data and saves it for later. The point is that even if the early results look bad, the car can change after software updates, so the “driver” or behavior you see later may be different.
Loss reserves are money an insurance company holds back to pay for claims later. The tricky part for AVs is that claims can take time to resolve, and software updates can change risk while the outcome is still unknown.
A risk transfer partner is the company that agrees to cover the financial risk if something goes wrong. They only do it if they believe they can price it correctly and still make money.
The judicial system is the legal process that decides who is at fault and how a claim gets resolved. Even if an AV system performs well, the final outcome can still depend on legal decisions.
Underwriting is how an insurance company decides whether to insure you (or a product) and how much to charge. It’s about judging risk, not just selling a policy.
An MGA is a middle company in insurance that helps write and manage policies for other insurers. The point here is that some MGAs push hard for growth, which can backfire if they don’t handle risk carefully.
The MG MGA is an older sports car from MG, built to be fun to drive. It’s usually a small, lightweight roadster, meaning it’s designed for open-air driving. People talk about it because it’s a well-known classic car that many enthusiasts still enjoy today.
Reinsurance is basically insurance for insurance companies. If a company expects big claims, it buys reinsurance so it isn’t financially crushed by those losses.
Private equity is money from investment firms that back companies and often push for fast growth. Here, the concern is that that pressure can make insurance risk management worse.
Robotaxes are autonomous vehicles operating as ride-hailing services without a human driver. The question “Who ensures them?” highlights a key insurance challenge: coverage, liability, and risk pricing for self-driving fleets.
Liability thresholds are the legal rules for when someone is considered responsible for harm. The host is saying the act doesn’t create brand-new responsibility levels, even though it changes how AV software is viewed.
“Risk follows title” is a liability principle meaning the party that holds legal ownership (title) is typically treated as responsible for certain risks and insurance obligations. For autonomous vehicles, the host is pointing out that even if software “takes over,” the vehicle’s registered/owned status still drives who must carry auto liability coverage.
If an insurance company pays for a crash, it may try to get that money back from whoever caused the problem. That “trying to recover” is called subrogation.
Cyber liability covers problems caused by digital attacks or software/security failures. With self-driving cars, that matters because the car relies heavily on software.
Loss ratio is an insurance math term that compares what insurers collect in premiums to what they pay out in claims. If it’s above 100%, insurers are paying out more than they collect, so prices tend to rise.
“Litigious” means a state or area where lawsuits are common. If lawsuits are common, insurance companies often have to pay more, which can raise premiums.
“Nuclear verdicts” are huge court-awarded damages in injury lawsuits. Even a few of those can make insurance much more expensive because insurers have to plan for worst-case outcomes.
Autopilot is Tesla’s system that helps with driving, like steering or speed control. It’s not fully independent—people are still expected to watch and be ready to take over.
“Human in the loop” means the driver is still part of the system. The car can assist, but a person is expected to watch and step in if something goes wrong.
Term
L4 survey V space
“L4” means the car can handle driving on its own in certain situations. The speaker is talking about how insurers think about the kinds of risks that show up for that level of self-driving.
When an insurer “settles,” it pays to resolve the claim without going to court. The idea is to reduce legal expense and avoid the uncertainty of a trial.
ODD means the “rules of where the car is allowed to drive itself.” Insurance pricing depends on how limited or broad that allowed area and situation set is.
Concept
insurance towers
“Insurance towers” means building coverage in layers so you can reach very high protection limits. For big risks, one policy layer isn’t enough, so they stack them.
Self-insuring means you keep money aside to pay for losses yourself. Instead of relying entirely on an outside insurer, you’re funding some of the risk internally.
A captive arrangement is when a company insures itself using its own insurance setup. Instead of relying only on outside insurers, the company controls how the risk is priced and funded.
A captive structure is the company’s internal insurance “system.” The company sets aside money for losses and can benefit if claims end up lower than expected.
Term
ODE's
This seems to mean autonomous operations that aren’t happening in everyday traffic. The speaker is saying those off-road/controlled deployments can still generate useful insurance data.
Concept
pivoted or expanded
They’re talking about companies moving into other kinds of automated systems beyond regular self-driving cars. The idea is that experience from one area helps insurers understand risk in another.
ADAS stands for Advanced Driver-Assistance Systems—features like automated braking, lane keeping, and highway assistance that help the driver but aren’t full self-driving. The transcript highlights that different companies define and brand ADAS capabilities differently, which complicates collecting consistent data for insurance underwriting.
It means the car can do some driving, but a person is still responsible for watching and stepping in if something goes wrong. That matters for insurance because it changes who’s considered responsible.
Adaptive cruise control is like regular cruise control, but it can slow down or speed up to keep a set distance from the car in front. Here, it’s part of the argument about how much you still need to watch the road.
“Takeover” refers to the moment when a driver must immediately resume control from automated driving or driver-assistance functions. The speaker argues that if drivers become less vigilant, requiring rapid takeover can be risky—even if the automation helps in some situations.
Term
FC
FC here is an acronym for a forward-collision safety feature category that’s discussed alongside automatic emergency braking. The point is that some forward-collision systems have clearer proof of safety benefits than others.
AEB is the system that can automatically brake if it thinks you’re about to crash. In the discussion, it’s brought up as an example of a driver-assist feature that has shown safety benefits in data.
A vigilance task is when you have to keep watching carefully for something bad to happen. The argument here is that “watching until something goes wrong” doesn’t necessarily make you safer if you’re not fully engaged.
Following distance is how much space you leave between your car and the one in front. Some driver-assist systems can help keep that gap consistent so you’re less likely to get too close.
Term
L2
L2 is a level of “partial automation.” The car can help with steering and speed, but you still have to watch the road and be ready to take control immediately.
“Keep lane” is the feature that helps your car stay in its lane. It can nudge or steer to keep you from drifting, but you still need to watch what’s happening.
An omission error is when you leave something out that you should have included. In AV insurance terms, it could mean the company didn’t account for certain situations or limitations. Those gaps can still create real safety risk even if nothing “obviously” fails.
This phrase describes a software-style culture of rapid iteration and experimentation. In the context of autonomous/automated vehicle safety, the host argues that this mindset can be dangerous because vehicle systems need rigorous safety validation rather than quick trial-and-error. It’s used to explain why some AV companies may be riskier than others.
Concept
human driving association
“Human driving association” is presented as a fictional or speculative organization tied to the idea of a future where human driving is restricted. It functions as a narrative device to discuss how people might react if autonomy becomes the default and human control is outlawed.
The speaker is discussing how insurance pricing could change if a “human driven analog vehicle” is considered riskier in a world where autonomous systems are demonstrably safer. The core idea is that risk-based pricing and market demand would make coverage for non-autonomous cars “fightfully expensive” (i.e., much more costly).
“Pure analog” here is a non-technical way of contrasting older, non-autonomous cars with modern automated systems. The speaker uses it to mean vehicles that rely on human control rather than autonomy/automation features.
Car
1967 Triumph TR4-A
The 1967 Triumph TR4-A is an old-school British sports car. The point here is that older cars weren’t designed with today’s crash-avoidance tech in mind, so they don’t “protect you” the way modern systems try to.
Cameras are sensors that help the car “see” what’s around it. The host is saying they can help prevent crashes, but if something gets hit, the parts tied to those systems can be expensive to repair.
Telemetry is the car’s recorded data from its sensors and systems. In a crash, it can help show what the vehicles were doing and when, which can make claims more accurate.
A GPU is a powerful computer chip that helps the car process lots of sensor information quickly. The host is saying that the car can use that processing to understand what was happening around it.
“Fraud quotient” is a way of talking about how often insurance claims involve dishonesty or arguments about what really happened. The host’s point is that more vehicle data can make those disputes harder to fake.
“Float” is the insurance company’s money it receives from customers before it has to pay for claims. While it’s waiting, the company can invest that money to help its finances.
This frames vehicle autonomy (AVs) as reducing driver stress and perceived risk, not just improving safety metrics. The idea is that when the car handles driving tasks, humans may feel less anxious about crashes or errors.
A “state insurance fund” is an insurance program run by the government. It exists so people can still get insurance even when private companies can’t or won’t cover them.
LIVE
Hello, and welcome to the Atonicast. I'm Kirsten Korosak, transportation editor at TechRunch.
And I'm Ed Niedermeyer. I'm the author of Ludacris, The Unvarnished Story of Tesla Motors,
and Elon Take the Wheel, available December 1st, pre-orders available now.
And I'm Alex Roy, the co-founder and general part of a new industry venture capital,
and the founder of the Human Driving Association, which will soon be making a spectacular comeback.
And no one could guess how that's going to happen. Let's move on. Today's guess.
Will it come back with human driving?
It will come back with human driving. I mean, human driving has a...
Guys, please.
It's gone away. How's it?
I ended on a dramatic note. Let's just digest it.
Okay, today's guest is someone fantastic who's going to answer all the questions
that the Luddite opposers of innovation have been posing as an excuse to why automation is not
inevitable. It's a man who, at the heart of ensuring automation, Steve Miller,
hub international, senior vice president of innovation. Hello, Steve.
Good morning.
He's ready. He's prepped.
I said we're going to ambush you with the crazy insurance questions. Let's just start with,
how does one become the senior vice president of innovation at a company ensuring automation?
What's your background?
Oh, man, that's a great question. I think you have to go to the genesis of my career in insurance.
I was actually asked recently, as a child, did I dream of this? A legitimate question,
which is funny, and the answer is no. There are very few people within insurance of the
financial industry that dream of it as children, and they're typically odd individuals.
But there's usually a familial tie. My mother-in-law was an insurance. I got put into it,
and that was 22 plus years ago. But specific to autonomy, by proximity to the Silicon Valley
and proclivity towards technology, I was always working in emergent tech. And in 2015, I had a
lunch with Carol Riley across the street from Stanford, pre-drive AI being formed and operating.
And at that point, drive AI was the seventh permitted company in the state of California
to test AVs. The first six were all OEMs, Tesla, GM, et cetera. So this was the first
company that had to go and comply with the DMV's requirement to find insurance.
The DMV, very prescient, very leading, said you need to have applicable insurance for $5 million,
and very smart. However, that did not exist. So I went back from that lunch. I started calling
insurance carriers and underwriters who all thought self-driving cars were really cool
as a concept, as a technology, and who almost uniformly laughed at me when I said,
great, I want you to take that risk and I want you to insure it. We were able to get one partner
to do it. We took that, and over the last 11 years have expanded it from a R&D application
for on-road testing through to every mode of autonomy, on wheels, on sea, in air, small delivery
bots, on sidewalks, up to class 8 tractor trailers. Are you guys the biggest insurer of AVs in the
world? Let's put it this way. If you want the biggest hamburger franchise, you go to McDonald's.
If you want the best, you go to In-N-Out. So I would say we absolutely have the most
track record, the most reputational clout, and the most expertise in the industry.
You really set him up for that, Alex. I didn't know where he was going to take it.
It's a real pressure cooker in here of questions. How do you insure AVs in the early days when
there's so little data and not many miles? The interesting thing is this is the highest of
technology, but it is so dependent on the interpersonal relationships. So this is what we
tell our clients, our prospect, and the industry rip marches, you have to spend extra time. The
bias towards founders sometimes is to do a couple of things. One is to think that
because they have something brand new as a technology, that it means a wholesale disruption
of liability and risk transfer, and it doesn't. The other is to think that there's an easy button,
and there's definitely not. So to your question, Alex, on how do you get early adopters,
you have to make the business case. You have to go to them, them being the insurance marketplace,
and say there's over a trillion dollars of TAM, of insurance risk transfer annually,
and 40% of that is auto related. So you have to say, okay, if that 40% in your profit area,
and profit is a loose word with auto insurance, it's not profitable, right? But if 40% of that
revenue comes from this type of risk, and this type of risk stands to be disrupted by this technology
in a decade, which is a blink of an eye, you need to pay attention.
So the early days was tough. And also early days, we were focused on, let's just get over the
threshold that the AV that's being tested with the human behind the wheel with the safety engineer
in the passenger seat is supervised, and is at least as safe as a human in that capacity.
And let's build up momentum and track record. And we get from there to growth, to experience,
to scale, to more market interest, to a competitive market landscape, and competition drives everything.
So that's making it easier. So now I'll ask my questions since I haven't gotten it in yet.
I want to go back to those early days. Drive AI was a company I wrote a lot about and later got
aqua hired by Apple for a little history lesson there. But when you were starting to build
just the basics, starting from this foundation of humans supervised, you then had to build models
off of that to talk about how to ensure once you pulled the driver out. I don't know if Drive AI
ever ended up doing that, but I'm sure there are other companies you ended up working with who did.
So when did that happen? And what did that require? Were you taking data from the data that
their specific fleets were collecting, and then also measure them against just driving data and
crash data and things like that? Or how did you build those initial models
to be able to ensure driverless, not just the human behind the wheel?
Yeah, I think a history lesson and a logistical administrative lesson on how the insurance
marketplace works is going to be the preamble to answer that question sensibly, which is to say
all these insurance carriers file their rates with each state. So you've got x number of
insurance carriers in the marketplace, hundreds or thousands. You've got 50 states, and you have
to file your rates based on the type of vehicle and based on the state, and then file the ability
to deviate that rate up or down by 30, 40, 50% based on individual risk characteristics. So
early days was, let's go to the marketplace, let's say, please invest in this space, understand
that you're going to want to be involved. And so the marketplace said, cool, we'll use the human
driven rate because it's all we have. It's all we've been able to calculate. And quite honestly,
it's the only rate that's worth the time and expense to go file. And let's just debit it
up as high as possible, or surcharge it up as high as possible, which is what they did.
Over time, data has been collected. The difficulty is that when actuaries look at
quote unquote credible data, they're talking hundreds of millions of miles, billions of miles.
And they're talking homogeneous data, which we don't have still in the space. We're still talking
about hundreds of thousands, millions of miles with different AV drivers. So it's really looking
back, been less of an exercise of collecting and crunching data that would be actually credible.
And more about gaining years of experience and starting to see how claims are going to come
across, how they're going to be adjusted, what sorts of improvements are made in the claims
adjusting process just by nature of having very intelligent onboard sensors and compute that
can actually help forensically determine a claim. And then what we're seeing now is just like you
have the waves of the troughs of disillusionment and the waves of hype in broader autonomy,
we're seeing kind of wave two of vendors in the ecosystem, experts that are coming in and saying,
okay, we're going to take simulated data combined with on road data combined with your safety case
specialists, we're going to blend that information into something that can plug into the actuaries
to help them really go for scale. I hope you didn't want short answers today. I mean...
No, not at all. I was just pausing because I know I really wanted to ask a question because
there's other follow ups I can ask, but we can circle back. So often it's easy to assume that
whatever works in sort of the early stages of a new technological adoption cycle is what's going
to work longer term. But what you're describing is really sort of a transition. And it seems to me
right because absent the scale of data that you need, in a lot of ways it's like transitioning
from treating these organizations as a person almost towards a point where you actually get
the data at the scale where you can treat it more like a traditional business. Is that early stage
of it? You mentioned safety cases and that is a situation where it is more like you're assessing
a human driver. Someone comes to you and says, I want a driver's license and they're drunk and
disorderly and can't dress themselves properly and can't string a sentence together. You're not
going to rush to give them the responsibility to control thousands of pounds of machinery at
high speeds. There's sort of a similar thing in this early stage, I imagine, with these organizations
as well. Can you talk a little bit about that? Ed, are you making the equivalency that some AV
companies are the drunk and disorderly in the room? Well, I would say we certainly see a variety of
attitudes in the space towards safety, towards operations. And we've seen companies come and
go because of those things. So are you asking, I want to understand what you're asking Ed,
are you asking how a company like what Steve's running assesses and ensures a variety of different
actors in a relatively new and emerging space like specifically autonomous driving?
And how you, is that what you're asking? I guess I'm curious about the bootstrap phase of it,
right? The qualitative versus quantitative approach because it's a bit of an art and
a difficulty on the insurance distribution process for AV codes is if they're having difficulty,
they're blind, it's a black box as to where the difficulty comes from. Is it my role?
And to back up, I'm a broker, right? My job is to work with my clients, you go out to the
broad market and find a risk transfer for them, right? I've actually seen some prognosticators
and people on LinkedIn that say nothing should be insured in autonomy because these are non-deterministic
AI outcomes and we don't have quote unquote aircraft level policies, procedures, SOPs in place.
Well, the answer is we have to go through this process. We can't make perfect the enemy of better.
And the reality is the first time that we insured a haul insurance on an airplane in 1911,
there were none of those things in place either, right? So it has to be iterative.
It hinges a lot on transparency, Ed. So there are the hard and fast numbers, right? It is,
okay, you've got a fleet of test vehicles, where are they located, what are their values,
how often they're driving, what time, what's the ODD, do you have the safety driver,
do you not, are you carrying passengers, etc. But it's also about honestly getting onto a
conference call with underwriters, letting them ask the questions, putting the right people in the
room, getting the CTO on the call, getting the operations team, getting the regulatory,
the attorneys involved, and then looking at risk transfer. So one of the things I think is interesting,
and I'm sure you guys are going to ask questions about Build America 250.
But one of the things that's interesting about that is that there's no provision in there that
mandates how liability works, because liability is very well developed and has been in automotive
for the last 100 years. So risk transfer is going to follow negligence and liability. And so
understanding how those plug-in also becomes critical. Just to clarify, I think a lot of people
imagine ensuring a visa, and again, what the end state will be, which is you're
purely measuring or almost entirely just measuring the performance of a particular system.
But in the early stages, that's not right. You also, it is, it's how much testing are you doing,
what are your protocols for safety in that? There are so many factors that have nothing
to do with the performance of the system itself that go into understanding what is the actual
risk here that drives the outcomes of whether or not they're claims, especially. But I mean,
you think of most of the ADS crashes that we've seen, I imagine a lot of the claims. It's the
vehicles being hit rather than the vehicles going, the system's making a mistake and then
hitting something else. So it's just fascinating to me that there's all of this rich other stuff
that goes into the outcomes that we're seeing now, and that fundamentally, that's very different than
where this eventually will all go. And so that transition just strikes me as really fascinating.
Perfect example, Ed. We have this idea that because the autonomous system is safe, that the losses
that are accumulating for these companies won't enter to them, won't stick on them, right?
But here's an example, right? You've got an AV that's driving appropriately, driving safely,
that to your point is hit by a third party, that third party's at fault, right? Now the AV is
damaged to the tune of a couple 100000 dollars. Okay, great. The other party takes care
of it from the insurance. Wait, hold on a second, not if you're in California, your minimum state
limits are 15,000, right? So now where does the loss go? It sticks right with the insurance carry
on the AV, and now that's an AV, quote unquote, loss that sits on their loss record. So it's not
a perfect solution because we're not going to go binary, non-autonomous, and then flip the switch
and everything is autonomous and acting rationally. The human variable is a tough part.
When companies are a go, but there are other companies that still exist, go through the process
of trying to get third party validation from too sued and says, oh, your safety driver training is
cool, X, Y. How much weight does that carry with insurers?
It carries a lot, if only as evidence of kind of the management qualifications and being good
stewards of the organization, right? So if you're putting the right people with the right titles
in the room to interface with insurance carriers, they understand that you're being a responsible
corporate citizen. They understand that you're not just giving lip service to the deployment,
simulation, close course on road with a safety driver, pull a safety driver. I mean,
we could all map this out by rote. So it really becomes qualitative in how you're presenting.
We've done this validation. Here is the information. But also, when a question is asked
about how do you operate safely, how granular are you getting? Who are you putting in the room?
How much time are you spending with the insurance carriers to make them understand your risk?
What I counsel to my clients, and I'm going to give this to the marketplace right now,
is you cannot over engage with your insurance carriers. You do not benefit from being coy,
from being combative, any of those things. They're not going to serve you long term.
You might win the battle. You're going to lose the war. You need to engage with the marketplace
because you're going to need them. And quite honestly, you're not going to destruct them.
We've seen all manner of insure texts come out over the last few years. They tend to be really
good at technology and a little short-sighted on the insurance side, as to how claims develop,
how you have to be well-capitalized, the regulatory burden. It's a hard industry. That's why it's
slow moving. So I'd say you cannot over engage with your risk transfer partners.
I want to get into some of the recent legislation in a minute, but before we do that,
you talked early on about how when you initially went out to the marketplace,
people kind of scratched their head. Obviously, it's evolved. I'm sure the insurance industry
is far more educated now, but there is a big difference in how companies are approaching
autonomy. It used to be a very clear line like, here are the people doing ADAS. Here are the
people doing robotoxys. It was very specifically different, but now you have a whole group of
companies that are taking what some might call the Tesla approach, which is iterating and getting
better over time, and then suddenly it's going to be driverless. We're actually seeing a shift
in the marketplace because of that. Has that changed how the insurance company is assessing
these companies? Because something could dramatically change in terms of the risk profile
within a singular software update. Yeah. I think that's a good point to my previous
comment. We're starting to see validation efforts within the industry that I've always said,
it may be easier for autonomous engineers and experts to learn and plug into insurance
and the other way around. Early on with Drive AI, we were hosting insurance carriers and they were
saying, can we have your data? The answer is, what data? Do you know what to do with it?
No. They have no clue. It's better, but I think that there has to be a bridge in that. There has
to be an actuarial and engineering expertise in the middle because, Kirsten, to your point,
the software, let's just say if you've got the AV driver and Waymo says it's the smartest driver
in the world, but then there's an update. Well, now it's a different driver. Right. Even if you're
going to say at scale, it's got 50 tickets over a month because it's on thousands of vehicles.
It might look like the worst NVR in the world, but obviously there's a lot of miles being driven,
but the moment it updates, it's a different driver. Now, hopefully better, but in reality,
better in some ways and probably learning in others. That's a really good point. That's where
being able to take simulated data combined with on-road data, combined with explaining the update,
you're right. It is going to have to be flexible to understand that that is a new risk.
It's not to that point of granularity yet. Do you see, though, eventually, because oftentimes,
these software updates, we might hear about them because of a recall that has happened,
or there's a big announcement or a step-up change. Waymo might talk about it in a positive manner,
but we're not seeing all the little mini software updates that are happening all the time.
Do you think that the insurance industry will evolve in such a way that it will be able to
be communicating with its companies, let's say Waymo or whatever the company is,
and they will have enough data to be able to adjust that coverage in real time, or is that just
because to your point, it is a new driver every time?
Yes, but it's also aspirational. Insurance works on actuarial. It works on historical losses
that are projected for future loss reserves and then building a layer of profit on top. I get the
joke or the complaint sometimes that insurances like Vegas, the house always wins. But unlike Vegas,
with insurance, you want the house to win. You need your risk transfer partner to make a profit,
otherwise, they won't take that risk. But that means they have to understand, and usually,
they have to wait a bit to understand how the claims play out, because it's not just the efficacy
of the system and the software. It's also the judicial system and how a claim is going to
actually be adjusted and settled in or judged on. So this is actually an interesting parallel here,
both the AV industry and the insurance business are businesses where
things can look great, and you can be more aggressive than everyone else, and it looks like
you're getting ahead of everyone else, and then the long tail catches up with you, and it turns out
you weren't ahead. I'm curious, we've seen this happen in the AV space. We've seen it not happen
in some cases as much as maybe some of us think it should. I'm curious, is this something that
is an issue in the insurance space as well? Because especially, I think, if you're the
opportunity to say, hey, we're the new disruptors, you mentioned new entrants in the space,
you have to imagine there's always going to be funding for the companies that
are going to be more aggressive with their pitch, sell the vision of more aggressive,
yeah, underwriting, and then they get caught on the backside. Do we see that?
A thousand percent. There's an MGA, which is managed general agent, every day of the week.
They come in with some algorithm, they find a reinsurance paper, they might be private equity
backed, which is kind of a dangerous role for an insurance provider to be in, because that's
grow at any cost, and grow at any cost means don't be responsible. What we typically find
is that the quote-unquote exit, the news blurb you see for the successful exit,
is they've failed and they're selling their software back into the traditional distribution
model, which is good. What I am starting to see is savvy founders come in and say,
we want to be a service provider to the traditional ecosystem,
because I'm not going to secure and defend traditional insurance. I know why it is how it
is, but how it is is slow, archaic, frictional. I mean, we're not on video. I'm bald. I wasn't
when I started this career. It is stressful and it needs to be improved, but I think that the
technology should come in and aim to be additive rather than disruptive.
Yeah. Edward, you seem poised to say something.
No, no, I'm waiting for you. Who ensures Tesla,
like the robotaxes that are out there? Who ensures them?
You know what? Great question. Great question.
I can tell you that rumors on the street, I don't have too much concern about upsetting
Elon, and it doesn't seem like Eric does either by the subtitle of his new book.
He's not a very insurance. It's not big, as might be said today. He's not a big fan of insurance.
Yet he launched an insurance product.
See my previous comments about the way insurance products can be deployed, but also,
but also there is there is truth to the fact that Tesla is an example, right?
As a as a fellow skeptic, just because the scope and ability of not sorry, sorry, Alex,
just because the scope that was promised by Elon five years ago was not capable five years ago,
I can't let that cloud my judgment that it may not be achievable now, right? The technology
is improving. I wouldn't say that just because I don't believe it's a responsible
insurance approach that's being taken today, doesn't mean market share won't be gained,
and it won't survive because it's well capitalized to become disruptive tomorrow.
But what we're seeing is that insurance program is essentially it's not actuarial based.
It's aspirational. It says if you're driving a Tesla, we're going to give you a 50% credit,
because we believe it's 50% safer than a human driven vehicle. I think that they're probably
right, right? But there's no basis for it beyond Edward, you know, that's called that's called
insurance narrative command. Okay, fascinating. I have so much to learn. I do want to I do. Well,
are you more? No, okay, because I do want to ask about the build America 250, but because I'm just
curious what what you're seeing out of that as it applies to autonomy and insurance, because
you mentioned it earlier. So interesting, because as it applies to autonomy and insurance, it doesn't.
It was intentionally omitted. In fact, it references back to the act specifically does
not set up new liability thresholds, insurance regulations, etc. It does mandate that the
AV software be viewed as as taking over the driving responsibilities, right? But we still
know risk follows title. So if you are a motor carrier operating in Texas, using a third parties
AV stack, you know, you're still registering the vehicle. How does that work? It's probably going
to be auto liability first. And then there's a product liability subrogation that happens if
the crash occurs because the AV set malfunctions, there may be cyber liability, technology,
errors and emissions, we can get deep dive into coverage. But the point is that the
legislation as proposed stays out of the liability framework, which goes right back
to what I was saying before. This industry is old. It is it is experienced. And it sits right
alongside the legal framework, which again, we have a century of understanding how OEMs and
tier suppliers work together through contractual and liability risk transfer, or just legal
requirements, and then the whole judicial system as to how these things are adjudicated.
So obviously it was intentionally left out. You kind of hint at the edges of why, but maybe
you could say it more explicitly. Why do you think that this was left out with the intention there?
I'm going to answer a question with a question. Is Congress qualified to write procedures on
how AVs should be certified? They don't have that expertise, right? So they are going to
require self certification. Very similarly, they don't have the expertise to sit there and give an
edict on how liability should sit. And the moment they try to do that on one permutation,
there's going to be a thousand others that are not going to fit the box. And it's going to be
instructive to the insurance landscape, the attorneys, the operators. I mean, it is a mess.
So as difficult and problematic as the judicial system, the legal system is
in the United States, it is well established and it is creaking along under the burden of its own
weight functionally. And so I think that the idea is stay out of that and let it figure itself out.
What is the role of states in all this? I mean, you would assume that there's been so much talk
about sort of trying to build federal frameworks so that there isn't this, you know, the patchwork,
right? How is that playing out in the insurance space? How different is it state to state? Are
some states much harder to insure than others? Yeah, it has less to do with autonomy and more
to do with their individual litigiousness, right? And or the administrations in each state and how
friendly they are or the ODD and weather. But, you know, there are states that are bad to operate in
because it's just expensive to have an accident and have a personal injury attorney get a hold of it
and then go get a settlement that's a nuclear verdict to the tune of billions of dollars or
hundreds of millions of dollars. It's one of the funny juxtapositions to me, right? Like, so here
in Oregon, where we both live, we don't have the giant billboards for personal injury attorneys.
What's fascinating to me is you go to Texas, Arizona, a place where the AVs are,
Florida. These are the places where you do have the billboards. I've always found that
kind of a fascinating juxtaposition. It's just a coincidence, right?
Yeah, there's no room for the personal injury billboards in Oregon because all the weed packs,
I mean, we have bigger industries here, yeah.
Yeah. I mean, it's one of the juxtapositions and, you know, just unintentional difficulties that AV
faces is the insurance marketplace from an auto standpoint hasn't made a profit since 2012,
aside from a few quarters during COVID when people were paying for premiums and they were at home,
they weren't driving. And ironically, that actually caught up to it because there was more
severity in 2020 on accidents than expected because the highways were empty and so people
were driving fast. And when they were getting into these less frequent accidents, they were
damaging, right? But the auto industry has been essentially running a 116% loss ratio
for a decade and a half. Every dollar of premium they pay in, they're paying out a dollar and 16
in claims and overhead. That means auto rates have continued to go up. We all pay our own auto
rates. It's painful every year. That's what commercial businesses have been experiencing as
well. And those are the headwinds that the AV companies are facing because they're coming in
to solve the problem, but they're jumping into the middle of the problem and they're being judged
and kind of brought along with it. So they have that headwind and then just regionally, right?
If you're going to go anywhere in the Southeast, you're going to pay more. If you're going to be
in Louisiana, you're going to pay a crazy amount because they've just got a litigious
state and they've got laws that lend itself too bad. So, and you mentioned nuclear verdicts.
I mean, certainly, and obviously, the Tesla situation is different with autopilot. We saw
that huge $230 million settlement. Obviously, Tesla clearly wasn't insured against that specific
risk. There was a human in the loop. There's a lot of complexity there that is different.
But we haven't seen that kind of verdict happen in the L4 survey V space. Certainly,
it seems like that potential might always exist. How do you think about that sort of
really extreme kind of tail risk? Not even necessarily that the incident itself may be
particularly catastrophic. It's just happening in the right place, the right time, the right
people and the right lawyer because we've seen lawyers take shots at Tesla and sometimes it's
the right one and the right case is what makes that breakthrough. How do you work that into
modeling the risk for this technology? I mean, you have to realize that the
inclination from an insurance carrier standpoint is to settle. And so, they're going to always,
it's not that we haven't seen any V accidents, it's that we haven't seen them go to trial
because insurance carriers are savvy and they know the cost to pay lawyers to defend,
to get in front of a jury to be sympathetic to an injured part. So, the first line of defense is
and then you get into the question is how much insurance should I buy and the answer is always
as much as you can afford. So, it is a calculation of what is your total fleet? What is your ODD?
Are you carrying a bus full of passengers? Are you running delivery of goods? So,
then you're talking about building insurance towers in the tens of millions. And then what
portion of that do you self-insure? What does your balance sheet look like? Are you private?
Are you funded? Are you in the public markets? Do you have the ability to take on that risk
yourself, right? Because we're talking about how does the insurance marketplace understand
Kirsten that the new quote-unquote software update is a new driver. There's a point in the
maturation process where scale is large enough and if funding matches that a captive arrangement
makes sense because the whole point of that is that the AV developer may know their risk
profile better than any third-party insurance carrier could. So, why not retain that into a
captive structure which is to say they are pricing and they're setting aside
loss funds in capital for their own risk, but then they're participating in the profits because
the thing about insurance is it's a very defined transaction. I pay you to take on my risk and
if I don't have losses, I don't get my money back, right? So, I think that there's a stepping
ground where until actual credibility comes to fruition, it might make more sense to take
their own risk. How long will it take to get to that point where we move away from aspirational
and we really have data in which we can definitively make assessments on all the claims that actually
a lot of autonomous vehicle companies claim today, which is that if this is broadly scaled,
we will see reduction in crashes or more frequent reduced fatalities.
How long, based on what we know right now, when will we get to a point where we move out that
aspirational bracket and into something more based on real data? I can confidently say more than a
year and less than 50 years. How's that? Here's the thing. When I started this in 2015, my first
born was nine years old and I was like, this is great. She'll never have to have a driver's license.
Well, she's going into her junior year in college. Not only is her car in the driveway, but so is
my 16-year-old son's and I'm sure my 10-year-old will also have a license, right? So, these things
take longer than we expect. The reality, though, is we're getting actuarial experience from ODE's
that are not strictly on the road, right? The entire, the rising tide is lifting all ships and by that,
I mean, every one of these A.B. companies that has pivoted or expanded to also do industrial
autonomy, to also do defense autonomy, to do any other mode, those learnings are being fed into
the ecosystem, right? And so insurance carriers are becoming more and more comfortable. And I think
there does become kind of the butterfly effect. It's just the mass of weight at some point is
going to tip the scales, but I can't tell you when that will be. What about ADAS because that one's
tricky now because, first of all, every company has a different definition. They've all branded
them differently. But if we were to take the most basic definition and say, hands off, but driver in
the loop on highway, okay, we'll be real specific, is there any historical data that has been collected
on that that you see that there's any takeaways that the insurance agents or the insurance industry
has been able to actually take away from that? Or is that data even being collected?
That's a great question. It trends itself more towards personal insurance, right? We all have
vehicles with some level of ADAS. Alex is going cross-country in his, I'm in a 2019 Audi, so I'm
using adaptive cruise control with full mileage that if I'm in a mountain road and the curve
is coming and I'm following a vehicle, may take the turn before me on the curve that I'm paying
attention because my car thinks it's time to go 80 miles an hour straight, right? So I know the
edge case is in my own system. I have an opinion on this, which is, and I think it's shared by my
underwriting community, working more on the AV developer side, which is if we've agreed that
humans are bad drivers and the evidence is in the deaths per year, we're bad drivers because
we're inattentive. How do we become a better risk if we're turning over operation of the vehicle
to the software, except when we need to intervene in a moment's notice, right? I don't think that
ADAS in that formulation is going to be a net positive for safety. Well, we still, we've been
waiting, right? The IHS still has not sort of said, right? They have with AEB with FC. I love,
I love that these systems people don't know about, right? That occupies 0% of our mental space.
Like the IHS came out years ago and was like, yeah, clear, unambiguous, in the data, no question,
there's a safety benefit here. We still haven't seen this for all two systems. I've been waiting,
like, you know, as a skeptic in these things, you have to get used to the fact that someday,
you know, maybe someone like IHS will come along and say, no, you've been wrong about this,
your intuition sitting there in a vigilance task and waiting for things to go wrong and jumping
into takeover is not improving safety. Like maybe the data will prove you wrong at some point. And
I've literally been waiting for IHS to come out and say that I am wrong about that. But like the
intuition, which I think you just described quite well is holding pretty firm as long as IHS isn't
coming out and saying, yeah, like the data shows that there is a safety advantage to this technology.
Now, it's situationally specific, right? Me in my car, do I think it makes me safer?
I'll be honest, right? I'm not a perfect driver and I'm distracted like everybody else.
A CDL in a tractor trailer that is not distracted that the L2 is helping manage following distance,
helping keep lane, helping do gas savings. Yeah, I think that could be beneficial.
But I think you're just your human driver in their passenger car. It just doesn't,
we don't have the standard of capability in a ditch. Well, and insurance, right? Because a car
is not cars and when we buy a car, it's not limited to certain ODDs. And so as insurance,
you can't, it's very difficult to be like, well, this system, there is a safety advantage in certain
ODDs and not other, well, there's just no safety advantage because you don't, it's a product that
is not tied to a specific ODD, right? So I totally agree with you. I mean, like here in Oregon,
the opportunities for, I feel like to use, to use ADAS are pretty slim, spending a few,
you know, a week or so in Arizona where there's just all of these freeway, you know,
especially in Phoenix areas, you know, five lane freeway. If all my driving was that,
then like, yeah, I could see using an L2 system a lot. But how big is the gap between like the
best and worst actors among AV companies who walk in the door looking for insurance?
Oh, and what constitutes like that gap? Like how many of these companies don't know
what they don't know? And it's just errors of omission versus we got any, we need insurance,
but let's see what freaking wing it. So you're asking basically like, what is it that is the
clue, immediate clue that maybe they, they're not the best doctor? Yeah, I think it can be pretty
large. I don't ever think it's intentional, but it's the Silicon Valley. And so you get that
move fast and break things mentality, which doesn't work with vehicular safety, right?
I think ego is the primary driver. And unfortunate that our team over the last,
you know, almost dozen years, we either by self selection or luck tend to align with companies
that are very good stewards of their safety. But I have run across those where, you know,
there's just this laissez-faire attitude or, or maybe even an easy button, like, oh, you know,
we have a certain price point that we're trying to go into the market with. And this has to fit
into that. And that's a business reality, right? My job is to go find value and value means the
right combination of partnership, coverage and price. Price is important. It's critical. And it
needs to be less expensive, right? That's the whole, the whole thing that's trying to be unlocked.
But, you know, people, founders, companies can maybe get too focused on the economies and,
and short, give short shrift to the engagement that needs to happen.
But again, I don't, I don't find intentional bad actors in the space.
That doesn't make it any better though.
I'm like, these people are criminals. All right, let's go to the fun part.
The human driving association, the future of human driving in a world of rising automation.
You've, you've heard this, the Rush song, Red Barketta.
I, you know what? I'm going to have to go spot it. I'm sorry.
Looks like a Rush guy. Okay. Well, in the future, basically, it's throughout a scenario in the
future where it's illegal. It's humans cannot drive because autonomy exists. It's demonstrably
safer by orders of magnitude. And a human driving is just outlawed and the guy takes
his old Ferrari out for a drive and he's being chased by the police. So is that so crazy? I mean,
it seems like in the future, I mean, it seems obvious to me in the future, autonomy will be
demonstrably superior. And then as a result, a human driven vehicle, eight ass or not,
let's assume there's old cars that are pure analog. They run and someone wants to take
their car out and just market for it seems to be obvious that market forces will dictate
that the insurance for a human driven analog vehicle in a world of demonstrably safer automation
is going to be fightfully expensive. True false. I'll tell you what I think. I have driven down
the freeway and as a kid that watched Top Gun way too many times and sat on the roof of my parents'
house watching the Blue Angels practice during, you know, airshow weeks, watching them do those
maneuvers and being told by the announcers how close they were and the speeds that they were
operating being awed by that as an adult driving down I five at 75 to 85 miles an hour,
eight or 10 feet from the car to the night to my right. I've had the thought, what in the hell
are we doing? Right? We're driving missiles down the road. It's inherently unsafe. What if we do
this the opposite? What if instead of our baseline, we say we have a right to drive? What if we go
back 100 years and tell people we're going to travel like missiles down the road right next
to other missiles and we're going to do it while fiddling with our radio and a smartphone and,
you know, glassing up at the road every now and then? I don't know. The more I get into
autonomy, the more I'm like, is this something that I really need to do? And I say it, Alex,
from a guy who has a 1967 Triumph TR4-A very nice to drive around and noodle around with, right?
That car is built, by the way, to survive or crash, not to allow me to survive or crash.
Well, and the flip side of this is also right. I mean, as you mentioned, the car insurance business
sucks. It doesn't make money. I'm curious, how much pressure is there? You mentioned
rising rates. Is that all that's going to happen on this? Is it just our insurance is going to get
more expensive? Is there a sense in the broader industry and the insurance industry that we have
to get towards more automated driving? Or we have to have ADAS that is developed in a way that does,
you know, really where the goal is not to make it look self-driving, but to really enhance,
you know, to improve the risk. Like, is technology going to be a major piece in fixing the car
insurance business? Or is the car insurance business just always going to be broken?
Oh, yeah. You know, there's so much more to it than meets the eye, right? So we think that cameras
are going to improve safety. Yeah, but they also turn bumpers from $500 into $5,000 if you bang
them, right? But one of the things we haven't talked about is fraud, right? So what I've loved
to see, and I have seen these in claims, are the little fender bender, where the person that gets
hit or is at fault, it doesn't really matter because if you're talking about like a tractor
trailer versus a sedan, you're usually going to bias towards the sedan being the quote unquote
victim. Well, not only do we have 360 cameras, but we have telemetry, we have the GPU,
relative speeds of everybody. We know when a tractor trailer, for say, would have started
breaking in a rear end scenario, and how long it took the human driver behind them to even pay
attention. The fraud quotient that he said, she said, that gets really frictional and really
expensive insurance goes away, right? It's very easy to adjust a claim. So I think if you control
this claim costs, you actually do control that tail and eventually you improve pricing. I also
want to be clear, I don't want to sit here and stomp for insurance carriers as entirely rational
market based actors, right? Like the reality is competition drives a lot of these things. The
reality is they invest the float between when they take in the premium and they have to pay the claims
and the reality is every time I see a rate increase, I also see an earnings call
with, you know, highest ever quarterly earnings. So I'm like, make it make sense.
Right. So yeah, okay, that's helpful context.
Yeah, jail for big insurance here, because that's also the bane of my existence.
Yeah. Alex? Oh, I've got a lot. We're going to run out of time before you on my question.
Well, I do have to, you know, move on with my day as much as I would like to talk about this.
I'll call you separately and get you on a podcast where we get into the nitty gritty.
This is just a warm up podcast. Okay.
Now, I'll ask one last question. Okay. So in the future, autonomy is emotionally safer.
There are people who want to collect old cars, the insurance skyrocket is great.
But for the people who don't want to buy a new car or just driving old cars,
there's going to be this interim period where they need to drive,
but they don't want to, to buy a car with autonomy.
And they've got to get from A to B. So what happens is they're like a state insurance fund,
the way they set it up, you know, in states where they have natural disasters,
people need insurance. How does that work play out?
It could be, I think, look, maybe this is just a sign of, of maturation and age, but
years go by quick and things change and they don't change in the big clip blink of an eye,
but they're, they're iterative. And I've seen that over the last decade plus in that scenario,
Alex, maybe the insurance actually for the human driver doesn't get more expensive,
because as you get more and more AVs on the road acting rationally, it almost acts as a buffer
to their buffoonery, right? If they're operating irrationally, but you've got nine out of 10 cars
or AVs and operating correctly, then maybe it saves them from their own negligence.
Also, cars wear out, right? Maybe it's 10 years, maybe it's 20 years. They don't drive forever,
and one in a hundred human driven cars is going to be a drop in the bucket. So
I think that you might actually see that go down longer term.
That's a perfect great end. That's a great answer. Thank you, Steve Miller,
Vice President of Innovation at Hub International. And thank you for joining us on another episode
of The Atomic Cast.
About this episode
Steve Miller of Hub International breaks down how autonomous-vehicle insurance actually gets built, starting with early California testing requirements and the reality that autonomy doesn’t automatically change liability. The conversation moves through underwriting inputs like ODD, fleet details, and safety cases, plus why insurers rely on supervised testing, simulated data, and large datasets. They also cover pricing mechanics (state-by-state filings, actuarial vs “aspirational” programs), software-update risk, claims outcomes, and practical broker/insurer engagement.
Steve Miller, SVP Innovation @ Hub International, explains one of the least understood but most critical aspects of autonomous vehicles: insurance. From the earliest days of self-driving startups like Drive.ai to today’s robotaxi deployments, Steve explains how insurers evaluate risk, liability, safety cases, software updates, and autonomous driving systems. Also: Tesla, Waymo, ADAS, AV legislation, trucking, fraud prevention, and the future economics of self-driving cars.
