Model 06 · Risk Anatomy of an AI Use Case

Most AI risk assessments collapse a complicated picture into a single score and a traffic-light colour — which is exactly how genuinely dangerous use cases slip through with a comfortable amber while harmless ones get held up at red.

This model breaks every use case into six layers — Identity, Data, Action, Autonomy, Reach, Reversibility — and tiers it by the maximum layer score, not the average.

Risk Anatomy starts with the observation that risk doesn't average. One extreme layer is still extreme, even when the rest of the scorecard looks calm — so a use case with a single 5 on Reversibility deserves Critical-tier handling, regardless of how benign the other five layers are.

The result is an intake conversation that surfaces real concerns before a line of code is written, and a governance regime that lets the harmless 90% move at the speed the business needs.

Download the model doc for the six-layer definitions, the scoring rubric, the MAX-vs-AVG worked example, and the original hand-drawn hypothesis sketch the model grew out of.