Insurance, at scale

Every human vertical carries insurance — life, health, property, liability. As AI begins to operate on its own, the vertical it creates will be insured too. Today, that market has barely been written.

1 cube = $10B

AI Materiality Terrain

The same AI interaction carries different weight depending on domain, authority, data access, and consequence path.

height = interaction materiality

AI Prevalence

From almost nowhere to almost everywhere: the share of each domain's activity held by AI, 2015 → 2030.

1 cell = 2pp · 2030 = forecast

01 · thesis

The record

AI liability is ungraded. What follows is an open standard for grading it.

North* builds and maintains the measurement layer: the registry, the materiality model, and the grade.

the dependency stack

 

02 · the object

The unit of legibility is the AI Role Registry: a source-anchored disclosure of every role AI holds inside a platform. Where it appears. What it touches. What it outputs. What authority it carries. Where each role lives in code — repository, path, function, provider call, pinned to a commit.

Not an opinion about whether a system is good or bad. A structured record of what is actually there.

the role node

 

03 · the gap

Every human vertical carries insurance. Life, health, property, liability: trillions in written premiums, priced against centuries of measured consequence. AI's consequence is accumulating now — and it is effectively unwritten. You cannot underwrite what you cannot read.

proposed instrument · coverage gap

 

04 · the grade

A registry makes grading possible. Each role resolves to a materiality: a function of domain, authority, data access, and consequence path. A summarizer in an internal dashboard and a recommendation engine in a refund workflow are different objects, and the grade says so — reproducibly, from source.

Platforms resolve to a profile: the composition of their roles, the distribution of their authority, the paths their outputs can travel. Two reviewers, same commit, same grade.

proposed instrument · grade plate

 

05 · the standard is open

The framework is public: the role taxonomy, the permission enums, the materiality model, the report schema. Anyone can produce a registry. Anyone can verify one against source. Underwriters price from it, buyers diligence with it, operators disclose through it. The method is trusted; the grader does not need to be.

proposed instrument · verification loop

 

06 · the trajectory

The open question is no longer adoption. It is authority. Firms using AI quintupled in three years. By 2028, an estimated fifteen percent of day-to-day work decisions will be made autonomously. Every one of those decisions lands somewhere: a customer, a transaction, a diagnosis, a hire. The standard has to exist before the claims do.

proposed instrument · claims horizon

 

07 · the line

AI is not measured by model type. It is measured by where it appears, what it does, and what happens next.

[specimen report] [request grading]