EU AI Act Compliance for Insurance: Underwriting, Claims & Fraud AI

Explicitly High-Risk

• Life insurance underwriting: AI assessing mortality risk, health status, lifestyle factors to determine premiums and coverage
• Health insurance underwriting: AI evaluating health risks for pricing and acceptance decisions
• Risk classification: Models that segment customers into risk categories affecting premium levels

Likely High-Risk (Case-by-Case)

• Claims automation: AI that approves or denies claims with significant financial impact on policyholders
• Fraud detection: AI flagging claims as potentially fraudulent — false positives can delay legitimate claims
• Property/casualty pricing: If AI pricing decisions significantly affect access to insurance (e.g., excluding high-risk areas)

Bias & Discrimination

Insurance pricing inherently discriminates by risk — that's the business model. The EU AI Act requires that discrimination is actuarially justified and not based on protected characteristics:

• Gender: Already prohibited in EU insurance pricing (ECJ Test-Achats ruling, 2012). AI models must not use gender as a feature or proxy.
• Race/ethnicity: Proxy variables (postcode, language, name) must be tested for indirect discrimination
• Disability/health status: For non-life insurance, health data usage is restricted. For health/life, it must be actuarially relevant.
• Testing approach: Fairness metrics across protected groups. Equalised odds, demographic parity, or calibration — choose metrics appropriate to the insurance context.

Human Oversight in Claims

• Claims above threshold: Define value thresholds above which human review is mandatory
• Denial decisions: All claim denials should involve human review — AI can recommend denial, human confirms
• Fraud investigation: AI flags suspicious claims; trained investigators make the determination
• Appeal process: Policyholders must have access to human review of AI-influenced decisions

Insurance AI governance must satisfy both the EU AI Act and Solvency II:

• ORSA: Own Risk and Solvency Assessment must now include AI model risk
• Actuarial function: Must validate AI pricing models — actuarial sign-off on AI-driven premiums
• Outsourcing: AI models from third parties (InsurTech vendors) must comply with Solvency II outsourcing requirements AND EU AI Act provider obligations
• Data governance: Solvency II data quality requirements apply to AI training data