Insurance AI Model Risk Management Cross Border | Moring
Cross-border insurance model risk management means one control set that satisfies multiple regimes: the NAIC model bulletin in the United States, the EU AI Act with extraterritorial reach, and DORA for Solvency II entities. The alternative, parallel country programs, multiplies cost while leaving seam gaps.
Short answer
Cross-border insurance model risk management means one control set that satisfies multiple regimes: the NAIC model bulletin in the United States, the EU AI Act with extraterritorial reach, and DORA for Solvency II entities. The alternative, parallel country programs, multiplies cost while leaving seam gaps.
How it works
Cross-border operation is the normal condition for large insurers, and it determines the shape of model risk management. A carrier writing business in the United States and Europe answers to the NAIC model bulletin and state examiners on one side and to the EU AI Act on the other, with DORA applying to insurers and reinsurers authorized under Solvency II. Building separate programs per jurisdiction is the intuitive move and the wrong one: it doubles cost and guarantees gaps at the seams.
The better architecture is a single internal control set mapped to each regime. Core controls, a living inventory with named owners, risk tiering by decision consequence, independent validation, continuous monitoring, approved-use limitations, and evidence retention, are common ground. What varies is the mapping: the NAIC bulletin reads them as governance and documentation expectations for insurer AI; the AI Act reads them through its risk categories and duties for high-risk systems; DORA reads them through ICT risk and third-party oversight.
Extraterritoriality deserves specific attention. The EU AI Act imposes obligations beyond the European Union, reaching providers and deployers whose systems' output is used in the Union regardless of where the organization sits. A US carrier whose scoring models touch European customers can therefore be in scope without any European legal entity, which surprises insurers that mapped their compliance footprint by incorporation rather than by data and decision flows.
Jurisdictional differences also shape model risk content, not just process. Insurance-specific analyses of AI governance note that fairness, transparency, and documentation expectations vary across regimes, and that multinational carriers need consistent baselines with jurisdiction-specific overlays. A rating model deployed in multiple markets should carry one validation record with regional annexes for local filing and fairness expectations, rather than divergent national copies that drift apart.
Organizationally, cross-border model risk needs a center of gravity. A group-level model risk function should own the control set, the inventory, and the standards, while regional teams own local execution and regulatory liaison. Evaluation guidance for insurance governance solutions emphasizes centralized visibility with local accountability, which is exactly the structure that prevents both duplicated effort and unowned regional models.
The test of the architecture is a regulator's request. When a supervisor in any market asks for the inventory, validation evidence, monitoring record, and approval trail for a specific model, the carrier should answer from one system within days. Carriers that achieve that have genuinely solved cross-border model risk; those that assemble answers per request are running parallel programs in disguise.
Frequently asked questions
Which regimes apply to a multinational insurer's AI model risk?
The main set is the NAIC model bulletin and state examination expectations in the United States, the EU AI Act's risk-tiered obligations in Europe, and DORA for insurers and reinsurers authorized under Solvency II. Rather than separate programs per jurisdiction, carriers should run one internal model risk control set, inventory, tiering, validation, monitoring, and evidence, mapped to each regime's specific requirements.
Does the EU AI Act reach non-European insurers?
Yes. The AI Act carries extraterritorial obligations, reaching providers and deployers whose systems' output is used in the Union regardless of where the organization is established. A US carrier whose scoring or pricing models affect European customers can be in scope with no European legal entity. Insurance compliance footprints should be mapped by data and decision flows, not by incorporation alone.
How should rating models be managed across multiple jurisdictions?
Maintain one master validation record with jurisdiction-specific annexes covering local filing and fairness expectations, rather than divergent national copies that drift apart. Industry analysis of insurer AI governance notes that fairness, transparency, and documentation expectations vary across regimes, so a consistent baseline with regional overlays keeps the insurance model's evidence coherent while respecting local requirements.
How should a carrier organize cross-border model risk management?
Give a group-level model risk function ownership of the control set, inventory, and standards, while regional teams own execution and regulatory liaison. Evaluation guidance for insurance governance solutions emphasizes centralized visibility with local accountability. The test: any supervisor's request for a model's validation, monitoring, and approval evidence should be answerable from one system within days.