AI governance in insurance.

EU AI Act

AI used in life and health insurance pricing and credit scoring is classified as high-risk under Annex III. Insurers operating in or serving EU customers are deployers subject to human oversight, monitoring, and transparency obligations for these systems.

Anti-discrimination law

Insurance pricing AI that uses proxy variables — postcode, occupation, purchasing behaviour — that correlate with protected characteristics creates indirect discrimination exposure. Regulatory enforcement attention to AI-driven insurance discrimination is growing across jurisdictions.

Prudential regulation

APRA, PRA, and equivalent prudential regulators are extending model risk management expectations to AI used in insurance underwriting and reserving. In the US, SR 26-2 (April 2026, superseding SR 11-7) applies to insurers with US banking operations.

Consumer protection

AI claims assessment tools that systematically undervalue or delay claims create consumer protection exposure. Several jurisdictions have seen regulatory enforcement action against insurer AI that produced discriminatory claims outcomes.

Discriminatory pricing via proxy variables

Insurance AI models that incorporate postcode, occupation, vehicle make, or similar variables can produce pricing outcomes that are statistically equivalent to pricing on protected characteristics. Courts and regulators have found this constitutes unlawful indirect discrimination even where intent was absent.

Claims AI with insufficient human oversight

Automated claims assessment tools that make or strongly influence settlement decisions without adequate human review have faced regulatory action and class actions in multiple jurisdictions. The human oversight requirement is not satisfied by a reviewer who accepts AI assessments without independent evaluation.

Fraud detection false positives

AI fraud detection systems with high false positive rates create customer harm and regulatory exposure when genuine claims are denied or delayed on the basis of algorithmic flagging that is not substantiated by human investigation.

Model drift in underwriting AI

Underwriting AI models trained on pre-pandemic data produced unreliable risk assessments during and after COVID-19. Insurers without active model monitoring failed to detect this drift before it affected pricing and reserving decisions materially.

Has your underwriting and pricing AI been assessed for indirect discrimination — specifically, have proxy variables been tested for correlation with protected characteristics?

For claims AI, what is the human review protocol, and are reviewers equipped to make genuinely independent assessments rather than ratifying algorithmic outputs?

Do you have active monitoring for distributional shift in underwriting AI — and what thresholds trigger model review or recalibration?

Have your AI vendor contracts been assessed against EU AI Act deployer obligations — including audit rights, incident notification, and change management terms?

What is your adverse action explanation process for AI-influenced underwriting declines — can you provide a compliant explanation to affected individuals?

Has your actuarial function reviewed the governance of AI models that feed into reserving and pricing, applying MRM principles?