AI Insurance Cover in 2026: What Underwriters Cannot Yet Answer

The hardest question in insuring AI is not “how much?” It is “which policy?”

Risk is migrating from physical to digital, and digital risks refuse to stay in their lanes. Picture a mid-sized manufacturer running an AI system that schedules production and orders materials. The model drifts. It over-orders one input, starves another, and a line runs on the wrong spec for nine days. Product ships. A customer’s assembly fails in the field.

Now open the client’s policy binder and try to answer one question: who pays?

Cyber will ask whether there was unauthorised access. There was not. The system did exactly what it was told, and what it was told was wrong. Tech E&O will ask whether professional services were rendered to a third party. The manufacturer was not selling advice; it was running its own plant. Product liability will look at the finished goods and ask whether the defect was in design or manufacture, and the honest answer is neither. It was in a scheduling decision three steps upstream. Property will ask what was physically damaged, and the real damage is nine days of correct machines making incorrect things.

Four policies. Four defensible reasons the loss belongs to someone else. That is not a coverage dispute at the edges. That is a client discovering, after the loss, that they bought four contracts and own none of the answer.

The one thing every client wants from an insurance contract is clarity. Right now, AI exposure delivers the opposite.

History says this resolves, and says how. Cyber insurance did not begin as a product. It began as silent cover already sitting inside property and liability wordings, until enough losses forced carriers to decide whether they meant to be on that risk. Some wrote it in. Most wrote it out. The exclusions came first; the affirmative product followed. AI cover will almost certainly take the same road, out of today’s wordings and into language of its own.

Which means the useful work is happening now, in the gap. Not “what does our AI policy cover,” but a harder question: on which existing policies are we already carrying AI exposure without having priced it?

The talent implication follows. Underwriters who think fluidly across once-clean niches, who can hold cyber, E&O and product in the same head and see where a single event lands across all three, are about to be the most valuable people in the building. Not because AI is new. Because the lines between products were always a convenience, and this is the risk that stops respecting them.