The gap isn't in the AI. It's in the oversight.

Every firm now runs on AI, whether or not it says so out loud. Associates draft with it, analysts model with it, support teams answer with it. The technology arrived faster than the controls around it, and faster than the insurance meant to sit behind it.

The exclusions are already here

Carriers have started attaching explicit AI exclusions to professional liability and E&O forms. The language varies, but the effect is consistent: when a claim traces back to AI-assisted work, coverage gets contested or denied. That leaves a gap precisely where exposure is growing fastest.

The instinct is to treat this as a technology problem, a question of which model was used, or whether it hallucinated. It isn't. The exposure is a human oversight problem: someone failed to supervise, verify, or govern an AI output in the course of professional work. That is a negligence exposure, and it is now being written out of the policies that used to cover it.

Why oversight is the right unit of risk

If you underwrite the model, you're chasing a moving target: capabilities change monthly. If you underwrite the oversight, you're measuring something stable and observable:

• Are AI outputs reviewed before they reach a client?
• Is tool usage sanctioned and logged, or is it shadow AI?
• Does the firm have a governance framework, and does it hold under pressure?

These are behaviors, not products. They can be measured continuously, mapped to a recognized standard like the NIST AI Risk Management Framework, and priced.

What this means for pricing

Governance posture becomes the rating variable. A firm with strong human-in-the-loop controls is a fundamentally better risk than one where anyone can paste client data into an unsanctioned tool, and its premium should reflect that. Underwriting stops being an annual questionnaire and becomes a live signal.

That is the thesis Axiom is built on: insure the oversight, not the algorithm.