It is not frequency. It is aggregation.
Thousands of insureds bought the same three or four underlying models. One failure mode is correlated across the whole book.
An underwriter who hears “frequency is unknown” hears a pricing problem. An underwriter who hears “correlated across your book” hears the thing that keeps them awake. The pull toward keeping someone talking is inherited from the models underneath, and the best intentions at the app layer do not remove it.
When the affirm-or-exclude moment comes, exclusion moves first, because exclusion is free. The affirmative product comes after, and it goes to whoever can measure the thing. Silent cyber ended with affirm or exclude. This one will too.
Insurers moved first, and they moved by refusing.
In January 2026 the Insurance Services Office introduced three optional generative-AI exclusion endorsements for general liability programs, each excluding liability arising out of generative artificial intelligence. More than sixty property-casualty groups have since filed to adopt AI exclusions of their own. W. R. Berkley went further in 2025 with an absolute AI exclusion for directors and officers, errors and omissions and fiduciary lines.
Carriers price risk with no loss triangle all the time, through exposure rating, judgment loads, sublimits and tight wordings. What is missing here is not the pricing craft. It is any observable, comparable measurement of how the insured system actually behaves.
A pre-loss signal where none exists.
Something that separates one insured from another before the claim, which is the only thing that lets an underwriter write a line they currently cannot price.
Comparability across accounts
The same measurement, the same way, from one submission to the next. That is the part a carrier cannot build in-house and the part a broker cannot fake.
A loss-control arm for a line you cannot staff
The same shape as The Acheson Group in product recall: the carrier attaches a specialist, the insured is assessed before the loss and supported after it, and the carrier gets expertise it does not have in-house.
The precedent
Tokio Marine HCC Specialty Group and The Acheson Group, September 2025. A specialty carrier attached a domain expert to its Product Recall offering, pre-loss and post-loss, to strengthen a line with exactly this profile. That is the mechanism. We are proposing it in a different vertical.
A sorting tool, not a loss-cost model
We cannot yet correlate behavioral scores to claims outcomes, because there are almost no claims. The argument does not depend on it: the standard of care and the documentation are what make the risk writable, which is exactly how food safety became insurable before anyone could prove a single case.
What we do not claim.
We do not claim carriers are excluding chatbots
Exclusion endorsements exist and are being adopted; whether a given book is excluding is a question for that book.
We do not claim movement on combined ratio
We disclaim the causal link and would then be billing a loss-cost improvement to the most numerate buyer in the room. A sorting tool, not a loss-cost model.
We do not certify or warrant
A report is a measurement. It does not say the insured is safe, and it never will.