Terrorism cover, quoted by a model: what Hiscox got right

In August 2024 Hiscox said it had gone live with “the London insurance market’s first lead underwriting model enhanced by generative AI”, built with Google Cloud and starting with sabotage and terrorism cover (Hiscox press release).

The headline was speed: a quote “in a matter of minutes”. In-scope risks are assessed using Google’s Gemini model, and the system “generates an email for the broker with pricing and other data already completed, ready for underwriter review.”

Speed is the least interesting part.

Read the notes to editors

At the bottom of the release sits a footnote. The model “will initially focus on the renewal of existing sabotage and terrorism risks from the US and Canada, and will exclude the metro areas of New York and Chicago.”

That one sentence contains three good decisions:

• Renewals first. The model starts on risks Hiscox has already written and knows, not on new business it has never seen.
• A named carve-out. The two cities where terrorism exposure is most concentrated stay with humans.
• A human sign-off. The output is a draft for the underwriter, not a bound policy.

Our view: the exclusions list is the product

In the language of the FCA’s Mills Review, this is Level 4, “Approver”: the AI prepares, the human authorises. It is the level where, by the review’s own assessment, accountability “remains clear” even as oversight gets harder.

Most AI underwriting announcements tell you what the model can do. Very few tell you what it is not allowed to do. Hiscox did, and that is the part other carriers should copy. A published boundary lets brokers know when they are talking to a model, gives the senior manager something concrete to defend, and creates a clear moment when widening the scope becomes a decision someone has to sign.

What to watch

Hiscox said it would explore “potential applications in other lines of business.” That is the right ambition. Our test for each expansion is simple: does the new line come with its own exclusions footnote, and does a human still sign before the broker sees a price? If yes, this remains the model for the London market. If the footnotes quietly disappear, so does the reason to trust it.

Kate Markham, Hiscox London Market CEO, said the aim was “freeing up our underwriters from manual tasks and allowing them to focus on more complex risks where human expertise is critical.” We agree. The exclusions list is where that promise is kept.