In Depth
An occurrence policy ties coverage to the date of the harmful event, not the date of the claim. If the incident took place while the policy was in force, that policy responds whenever the claim eventually arrives, even after the policy has expired or been replaced. In effect, each policy year locks in permanent coverage for everything that happened during it. The policy can lapse, but the protection for that year's incidents does not.
That permanence is the appeal. The catch is that occurrence coverage asks the insurer to estimate, at the moment it writes the policy, how many claims might eventually emerge from incidents that have not surfaced yet, sometimes for years. For risks where harm reveals itself slowly, that long, uncertain tail makes occurrence forms expensive or simply unavailable. It is why most professional and technology liability lines are written claims-made instead, letting the insurer price a known, near-term reporting window rather than an open-ended one.
The practical difference for a buyer comes down to continuity. Under an occurrence form you do not need to keep buying tail coverage to protect old work, because each year stands on its own. Under a claims-made policy you do. Which form you hold is the difference between knowing your past work is covered and only assuming it is.
What It Looks Like
Say an IT services firm held an occurrence-based liability policy for 2020 (illustrative). During that year a project it delivered introduced a latent flaw, but the client does not discover the resulting loss and file a claim until 2023, three years and two renewals later. Under an occurrence form, the 2020 policy responds, because the incident fell within its period. It does not matter that the policy expired in 2021 or that the firm later switched carriers.
Run the same facts under a claims-made Tech E&O policy with no tail purchased, and the answer flips. With no active policy in 2023 to receive the claim, there is nothing to trigger. The incident and the harm are identical; the outcome turns entirely on which trigger the policy uses.
Why It Matters For AI Vendors
AI failure modes are often slow to surface. A biased pattern, a drifting model, or a bad output buried in a record can take a long time to ripen into a claim. An occurrence form would, in theory, cover all of it with no tail to manage. But that same long tail is what makes occurrence coverage hard to price, which is why AI coverage in this market is written claims-made rather than occurrence. Either way, check which trigger your policy uses instead of assuming it behaves the way you would prefer.