Library
White paperMichael Clark
Pricing AI Risk
Description-based pricing fails for AI estates on seven structural features of AI systems, so carriers should rate each governed system on peril-specific exposure, with hazard held apart from control state, using evidence a properly run governance process emits as it runs. The method covers the underwriting process, six contract structures, the insurance law limits and an indicative actuarial structure, with a synthetic demonstration labelled a proof of method. No rating parameters are stated, and the loss definition, trigger and monitoring position should be settled first.
Key findings
- Description-based pricing fails on seven structural features of AI systems, so an AI estate should be rated per governed system and aggregated to the policy.
- Machinery breakdown, not liability, is the nearest reference class for the rating unit, while cyber is the better model for accumulation through shared components.
- The loss definition, the trigger basis and the monitoring position should be settled before any rate, because none of them needs data.
In the paper
- Why the description fails
- Rate the governed system
- Three decisions come before any rate
- Make the submission sort itself
- Contract terms on facts the carrier can verify
- What a pricing committee should hear
Michael Clark
Chief Executive Officer, Standard Intelligence
Your next step