Strict Liability Laws Will Fix AI Insurance | Rune Kvist artwork

Strict Liability Laws Will Fix AI Insurance | Rune Kvist

MTS

October 18, 2026

AIUC CEO Rune Kvist explains how AI insurance works, why catastrophic risks can be audited without insuring total losses, and why novel agentic failure modes like loss of control and autonomous cyberattacks are changing frontier liability. Turn ideas into software people love.

0:00

Rune Kvist

There's a bill put forward that when an agent causes damage, someone should be held liable.

It can't be no one. Patching the current liability system is an extremely effective lever for creating the clarity that unsubmitted insurance markets both thrive on but also help create by kind of pushing the frontier of liability because that's the core of their business.

0:24

SPEAKER_2

And we are back. We are live with Rune Kvist, who is the co-founder and CEO of AIUC, the AI underwriting company, which there has been a lot of talk about insurance for AI. How will insurance work given advanced AI systems? How will we even ensure the risks of the current systems that we have?

And AIUC is actually doing it. So I've been very excited to have you on for a while now. Rune, welcome.

0:51

Rune Kvist

Thank you. Thanks for having me.

0:53

SPEAKER_2

So tell us about what AIUC actually does. Explain to the audience simply, how does it apply insurance to AI?

1:02

Rune Kvist

Yeah. So AIUC, we do two things. First, we build standards that you can audit the AI systems against. And then secondly, we write insurance contracts that cover the damages from AI.

The goal of these is, the question we start from is how do you build confidence that a particular system can be deployed?

1:22

SPEAKER_2

Interesting. So how do you do that?

1:24

Rune Kvist

Well, you start from a list of what are all the things that can go wrong. So you start from incidents that have happened or threat models of things that you think might happen soon.

From there, you define what are the controls that you want to have in place that could prevent those incidents. That's basically what a standard is. And you can think of a standard as kind of an underwriting framework that says, these are the things that need to get done to prevent these incidents. Then you go and audit against that standard.

Auditing here both means like, do they have the technical controls in place that they say they have? And secondly, are they effective? So that's where EVOS come in, Red Teaming comes in and saying like, great, if they have guardrails in place, are they actually effective? And you do that through thousands of simulations. And then you take the findings and you put them into an underwriting framework, which is where you come at pricing for insurance. And roughly what you're trying to answer is, what is the frequency of damages and what is the severity of those damages? And that's what comes together into an insurance price.

2:27

SPEAKER_2

Right. So to my understanding, pricing insurance accurately involves understanding a lot about the world, understanding a lot about your systems and the nature and scope of their risks, which is something that we have a lot of data on for automobiles, which is why there's such a robust car insurance industry with very standard pricing. But for AI, we kind of don't understand the scope and nature of these systems and of the risks which are also changing all the time. So how do you go about pricing AI risks and insurance?

3:04

Rune Kvist

Yeah, the first point to start understanding is that there's always someone underwriting an AI deployment, regardless of whether it's an insurer or not. There's always someone making a decision about, do we think the benefits will outweigh the cost? And so it is extremely hard to figure out what the risks are. But everyone is kind of implicitly doing all the time. Regulators are doing it when they're allowing a model to be released. Frontier Labs are making these decisions when they're releasing a model. And so the insurers have the same task, which is a hard one.

And you can go back to every, today we have a very kind of robust understanding of how cars fail. The first car insurance policy did not have that. They did not have much data and they did not have the adequate tests to figure out what that is. And so every insurance market that's robust today, they start out being not very robust. And the way you learn is through trial and error.

And, or two ways, trial and error. So over time, you'll accumulate more data on how the AI systems fail. Even if the frontier continues to progress, each model has a bunch in common with the previous generation.

And then two, what you have to start building over time is an actual kind of model of how the system works and where the risks come from. So when we do hurricane insurance today, we both have historical data on hurricanes, but we also have physics models of how the weather works. We don't truly understand how the weather works, but it turns out that these models can be a lot better than not having a model. And we'll be releasing some work soon on what does catastrophic risk models for AI look like. How do you take the tools that actuaries and underwriters have built over the last 20, 30, 40 years and apply them to AI? Because it does turn out that even in places like rocket launches or sometimes strategic war decisions, where there's not a lot of historical data to go on, you can still outperform human intuition with structured risk models that you can go and critique.

Read the other 20 minutes of this episode here.

$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.

Not what you expected? Full refund within 14 days.

Thousands of transcripts fetched by people building searchable podcast archives

Fetch transcripts with code

The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:

request
curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.

Using your own key:

request
curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000796939246