AI-IP RISK SCORING
The actuarial foundation that GIPL insurance pricing runs on. The metric the Underwriter Certification track is calibrated against and the scalar summary that portfolio reporting and premium-factor calculation reduce to.
The score formula
The AI-IP risk index combines mean failure probability E[p] and uncertainty σ: round(0.7 × E[p] × 100 + 0.3 × σ × 100 × 2). Seventy per cent reflects the central estimate; thirty per cent reflects uncertainty. The Premium Factor is current score ÷ baseline score.
Five exposure categories in E[p]
Training Data Dividend exposure, NILP Downstream Obligation exposure, Output Derivative Liability exposure, Pipeline Compliance Failure exposure, and Systemic correlation exposure. Each independently assessed from operator-specific data.
Variance-narrowing controls
Published cip.md: mean -15pp, sigma ×0.85. CDR registration: mean -10pp, sigma ×0.80. Vendor Representation Clause v1.2: mean -8pp, sigma ×0.85. CIP-certified Practitioner deployment: mean -5pp, sigma ×0.90.
Annual recalibration
Triggers: legislative change (UK DUA Act, EU AI Act implementing acts, US state NIL statute updates), AI capability change (new model architectures, voice cloning at production quality), and claims pattern change (documented frequency and severity changes from underwriter community).
CIP AI-IP Risk Architecture v3.7, https://creativeip.org/cari