The AI risk measurement problem

The original Inference Re draft argued for an exposure-measurement and evidence layer for AI liability. Keep that as a product-development thesis. It does not establish that all AI risk is unpriced, that no insurer has relevant data, or that every existing assessment is inadequate. GROUNDED · Brochure

Define, measure, control, verify

Start with a consequential workflow and a loss scenario. Define the eligible events, measure their population, identify the relevant control, and verify what occurred within the observed boundary. Then ask whether that information improves a specific insurance decision. STABLE

AI-related exposures can arise in cyber, Tech E&O, professional liability, employment practices, crime and other covers. Treatment depends on the actual policy and loss, not the “AI” label. Exclusions, silent coverage and dedicated cover are questions to inspect in wording rather than universal market descriptions. STABLE

Current programme and proposed products

Glacis’s Lloyd’s Lab selection is verified in the market-lab module. Specific alternative-risk-transfer products, wording, pricing, capacity and commercial adoption remain proposals unless separately established.

The academy explores whether bounded runtime evidence can support risk selection, loss-control review, claims investigation, and carrier-owned product design. Legal burden-shifting, safe-harbour effects and “first in the market” claims require primary legal or market evidence; the draft brochure supplies none of those conclusions. STABLE

Next

→ The missing denominator

Drill this page →23 bank questions stand behind what you just read. Check it while it’s warm.