AI liability in casualty: a global roundtable

Executive summary

Artificial intelligence has become one of the defining casualty conversations of 2026. The biggest question facing the market is no longer whether AI creates liability, but where that liability ultimately lands, and under which line of business it gets paid.

Howden Re's global casualty leaders met to discuss how this is playing out across the United States, the United Kingdom and Continental Europe. Their conclusion was clear: legal and regulatory approaches differ by region, but the underlying challenge is shared. The same AI-driven loss could become a casualty, cyber, professional indemnity or product liability claim depending on the facts, the policy wording and the jurisdiction involved.

AI is creating new opportunities for the market as much as new exposures and new protection needs. Carriers that treat it as only one or the other risk missing the other half of the picture.

The US is leading the push toward AI-specific exclusions, though softening market conditions have slowed widespread adoption. London remains open to underwriting AI-related exposures, with many claims still resembling conventional negligence rather than an entirely new class of risk. Continental Europe faces perhaps the most significant structural shift, as the new EU Product Liability Directive 2024/2853 broadens amongst other things the definition of a product, the definition of damage and the range of parties who could ultimately be held liable.

For carriers, this is becoming less about predicting individual AI claims and more about understanding aggregation, wording and cross-line exposure before the market sees its first defining legal precedents.

To explore what this means for the market, Howden Re's Josh Everdell, Head of Global Casualty, Carrie Byler, Head of US General Casualty, George Harris Hughes, Managing Director, Global Specialty, Casualty Treaty, and Wolfram-Ferdinand Schultz, Head of Casualty for Continental Europe, share their perspectives on where the market stands today and where it is heading next.

Is AI creating genuinely new casualty exposure?

That question sits at the heart of today's market debate.

"The bigger question is whether this is a completely new exposure or existing exposure with a new name," says Josh Everdell, Head of Global Clients and Head of Casualty, Howden Re. "Where we see movement around the exclusions is largely on the excess and surplus lines side."

George Harris Hughes, Managing Director, Global Specialty, Casualty Treaty, Howden Re sees a similar discussion playing out in London.

"A lot of what gets called AI liability in London is really just negligence with an AI label on it. If a professional relies on an AI tool and produces flawed work, that is a standard professional indemnity claim, not a new species of risk."

The distinction matters.

Why are markets responding differently?

Although AI is a global issue, the market response remains highly regional.

United States

"In the US, we’re monitoring the trends," says Carrie Byler, Head of US Casualty, Howden Re. "Carriers are evaluating the various coverage forms and determining the most appropriate way forward."

For many current claims, AI is changing how established liability scenarios arise rather than creating entirely new legal principles. That means insurers cannot simply ask whether they write AI risk. They need to understand how existing casualty portfolios may already contain it.

United Kingdom

London's approach reflects a balance between regulation and underwriting appetite.

Where London may distinguish itself is through product development rather than regulation.

"London's ambition to establish itself as a global AI hub isn't just about attracting technology businesses. It's also about developing creative underwriting solutions for this exposure."

That's already happening. Apollo ibott, the Lloyd's syndicate known as 1971, recently brought a dedicated AI liability product to market. It offers affirmative general and product liability cover for AI developers. It also has a separate solution for AI users whose existing policies may not respond to AI-related losses. The product itself suggests that London's underwriting appetite is starting to show up as real capacity, with other syndicates likely to follow.

Continental Europe

Continental Europe faces perhaps the most significant structural change.

From 9 December 2026, the new EU Product Liability Directive 2024/2853 will classify software, including AI, as a product.

That seemingly technical change could have far-reaching implications.

"We saw a case where a lawyer used AI to draft a legal brief that cited judgments which simply did not exist," says Wolfram-Ferdinand Schultz, Head of Casualty Treaty for Continental Europe, Howden Re . "The case was lost once the judge checked the citations."

"But that raises the bigger question. Is it the designer, the developer, the seller or the user who becomes liable? Once software is treated as a product, that question moves from academic to central."

The Directive also broadens potential reporting periods for complex claims and may extend liability into areas such as psychological injury in jurisdictions where legal precedent already exists.

Is a standalone AI product on the way?

Today, most AI-related exposure is still being addressed through existing liability frameworks. But as autonomy increases and reliance on AI-generated advice grows, that may not hold, and demand for more specialised, dedicated solutions is likely to build over time.

Commentary from across the market points to a similar conclusion: silence on AI is not a stable position. Most commercial policies do not yet define AI at all, which leaves insurers exposed to the basic promise to pay if a claim arises and the wording offers no clear answer.

1 January 2027 is emerging as a natural pivot point. A growing number of insurers are expected to introduce generative and agentic AI exclusions for general liability, and that clarity is likely to sharpen client demand for real, defined AI cover rather than close it down.

There is also a competitive threat building from within. AI developers are hiring experienced risk managers and building out sophisticated risk functions of their own, drawing on the same talent pool insurers rely on. Companies with large balance sheets have the means to build captives or front their own cover rather than wait for the market to catch up. 

Many companies using AI as a decision-support tool believe their exposure is unchanged, since a human still makes the final call, making any resulting claim one of human rather than AI negligence. As reliance on AI-only advice grows, that reasoning may not hold, and that shift may be the point where demand ultimately surges.

What should carriers be thinking about now?

Josh Everdell commented: “One feature of today's market stands out across all regions and that is that coverage decisions are moving faster than liability doctrine. That is not to say that there is a market weakness but rather reflects the reality that insurers must underwrite long before courts establish consistent legal precedent.

The challenge therefore shifts towards aggregation.”

Howden Re's recent research into AI-related D&O exposure highlighted how a relatively small number of widely used AI models could generate correlated losses across many insureds simultaneously.

The casualty market faces a similar question.

A court ruling, regulatory intervention or failure involving a widely deployed AI model could produce multiple claims across different industries, policies and classes of business rather than isolated losses affecting individual insureds.

That concentration risk deserves as much attention as the underlying liability itself.

Outlook

The regional differences that characterise today's market are unlikely to remain for long.

US carriers are expected to move from monitoring AI towards actively pricing it as January renewals approach. Europe's Product Liability Directive will begin testing entirely new liability questions from December 2026. London's underwriting appetite will eventually be tested by genuinely novel AI claims rather than traditional negligence cases involving new technology.

The legal answers will develop jurisdiction by jurisdiction.

The underwriting challenge is already here.

As with the wider reinsurance cycle, the challenge is not predicting exactly how AI risk will evolve, but helping clients understand and manage emerging exposures before they become material.

Carriers that wait for settled case law before reviewing aggregation, policy wording and cross-line exposure may find the market has already moved ahead of them.