The casualty reinsurance market already carries AI exposure. The question is whether it knows where.
That is the practical challenge running through a Howden Re roundtable published at Monte Carlo. Four of the broker's global casualty leaders mapped how AI-related liability is emerging across the US, UK, and Continental Europe. The same AI-driven loss, they concluded, can land under different lines depending on the facts, the wording and the jurisdiction.
Howden Re's review of current wordings found that most commercial policies do not define AI at all, which leaves underwriters with a basic promise to pay if a claim arises and no clear answer in the wording. Josh Everdell, head of global clients and casualty at Howden Re, framed the core question: "The bigger question is whether this is a completely new exposure or existing exposure with a new name."
Much of what is currently labelled AI liability resolves as a conventional claim, according to George Harris Hughes, managing director of global specialty casualty treaty at Howden Re. "In London, a professional relying on a flawed AI output still faces a standard professional indemnity case," he said. "The AI element changes how a loss is generated, not the underlying legal principle."
In the US, carriers are monitoring rather than acting. AI is shifting how established liability scenarios arise, instead of creating new legal principles. Howden Re's view is that the distinction is material for treaty structures: a cedant's existing casualty portfolio may already contain AI exposure, priced and reserved as something else.
Continental Europe faces a harder shift. From December 9, EU Directive 2024/2853 classifies software, including AI, as a product under strict liability rules. That reclassification extends liability beyond traditional manufacturers to software developers, sellers and distributors. The directive broadens what counts as damage and widens the pool of potential defendants.
Wolfram-Ferdinand Schultz, Howden Re's head of casualty treaty for Continental Europe, asked: "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 replaces a framework governing European product liability since 1985.
The deeper question is how much AI exposure is correlated. "Coverage decisions are moving faster than liability doctrine," Everdell said. "The challenge therefore shifts towards aggregation."
A failure involving a widely deployed AI model could generate claims across multiple industries and policy classes simultaneously. Howden Re's own research on AI-related D&O exposure found that concentration in a small number of widely used models creates correlated loss potential. Howden Re argues the casualty market faces the same structural problem.
Howden Re expects the January 2027 renewal to be a near-term pivot, with many carriers introducing generative and agentic AI exclusions across general liability lines. When that exclusionary language proliferates, demand for defined affirmative AI cover will sharpen.
Capacity is already forming in London. Chaucer and Armilla AI launched a standalone third-party liability product in 2025 covering AI hallucinations, model drift, and similar failures. The regional legal answers will develop jurisdiction by jurisdiction.
What the market shares across all three regions is the same underlying problem - claims will arrive before the wording answers them.
