10 September 2026 / 10 min read
Analytics team - Global RE Risk Science

Where it already earns its keep
The biggest wins so far are in the dull parts of the job. Pulling data out of submission packs, cleaning and geocoding schedules, spotting what is missing, and enriching exposures with satellite imagery and third-party data used to eat whole afternoons.
Handing that work to software gives underwriters more time for the parts that need judgement: structure, wording and price. Brokers notice the difference too, because they get a clearer answer sooner.
A person still signs
We treat AI output as a second opinion, not a decision. Every risk we bind is reviewed by an underwriter who is accountable for it, and any model that touches pricing or selection is documented, monitored for drift and checked so it cannot stand in for factors that have no place in an underwriting decision.
Regulators are heading the same way. In the US, a growing number of states have adopted the NAIC model bulletin, which expects insurers to run a written AI governance programme and makes clear that buying a model from a vendor does not hand off the responsibility. In Europe, the AI Act treats some insurance uses as high risk. Reinsurers will increasingly ask cedants to show this kind of discipline as well.
AI as a source of claims
The other half of the story is exposure. Policyholders are putting AI into customer service, lending, diagnostics, design and software, and a mistake can repeat across thousands of decisions before anyone spots it.
Most professional indemnity, D&O, product liability and cyber wordings were not written with that in mind. Research from the Willis Research Network this year found AI exposure building up across liability lines with nothing in the policy that addresses it. It feels a lot like silent cyber did ten years ago, and we would rather not learn that lesson twice.
What we ask cedants
We want to know how a cedant spots insureds that lean heavily on AI, whether its wordings affirm, limit or exclude AI-driven loss, and whether it has thought about many insureds relying on the same model or vendor.
It is cheaper to be clear now than to argue about it in a claim later. Cedants who deal with AI openly in their guidelines and wordings will find reinsurance easier to buy.