Insurance AI Model Risk Management EU AI Act | Moring
The EU AI Act, Regulation (EU) 2024/1689, is the first comprehensive AI law and applies risk-tiered obligations to insurers, including carriers outside Europe whose systems' output is used in the Union. Insurers should classify AI uses by risk tier and attach data governance, oversight, logging, and evidence duties accordingly.
Short answer
The EU AI Act, Regulation (EU) 2024/1689, is the first comprehensive AI law and applies risk-tiered obligations to insurers, including carriers outside Europe whose systems' output is used in the Union. Insurers should classify AI uses by risk tier and attach data governance, oversight, logging, and evidence duties accordingly.
How it works
The EU AI Act, Regulation (EU) 2024/1689, is the first-ever comprehensive legal framework on artificial intelligence, and it organizes compliance by risk. Commentary summarizes four tiers: unacceptable, prohibited uses; high risk, carrying heavy compliance obligations; limited risk, with transparency requirements; and minimal risk, with no specific obligations. For insurers, the framework converts model risk management from a best practice into a mapped legal duty wherever the Act applies.
Insurance use cases sit on different rungs of that ladder. Analysis of the Act's impact on insurance notes that AI is transforming the value chain, enabling personalized services and pricing insights, while regulatory obligations concentrate where systems affect consumers materially. Life and health underwriting, pricing and risk classification, and claims decisions are the natural focus for high-risk classification analysis, while internal summarization or search tools generally fall into lower tiers.
Obligations follow the role an organization plays. The Act distinguishes providers, who develop systems, from deployers, who use them under their own authority. An insurer deploying a purchased high-risk system still carries deployer duties: human oversight, input data relevance, monitoring, and logging. This is why vendor sourcing never removes model risk work, and why carrier inventories must record, for each system, whether the insurer acts as provider, deployer, or both.
Extraterritorial reach matters for global carriers. The Act's obligations extend beyond the Union to providers and deployers whose systems' output is used in the EU, so a North American insurer whose scoring or pricing models touch European customers can be in scope without a European legal entity. Compliance footprints should be mapped by decision and data flows, not by where entities are incorporated.
For model risk management specifically, the Act's high-risk duties read as a control checklist: data governance and quality for training data, technical documentation, record-keeping, transparency to affected persons, human oversight design, and accuracy, robustness, and cybersecurity. Insurers can satisfy much of this with the controls they already need, validation evidence, monitoring records, and oversight logs, provided those controls produce artifacts in the form the Act expects.
The pragmatic posture for insurers is to treat the Act as a tiering and evidence specification layered onto model risk management: classify each AI use into a risk tier, apply the corresponding duty set, keep the evidence machine-readable and current, and revisit classifications as the Act's implementation evolves. Carriers that do this once, centrally, avoid re-deriving compliance per market.
Frequently asked questions
What is the EU AI Act in brief?
The EU AI Act, Regulation (EU) 2024/1689, is the first comprehensive legal framework on artificial intelligence. It classifies AI systems into four risk tiers, unacceptable, high, limited, and minimal, with obligations scaling accordingly: heavy duties for high-risk systems, transparency for limited risk, and none for minimal risk. For insurers, it converts model risk management from best practice into mapped legal duty where the Act applies.
Which insurance AI uses are likely high risk under the Act?
Uses that materially affect consumers draw the scrutiny: life and health underwriting, pricing and risk classification, and claims decisioning are the natural focus for high-risk analysis, because they determine coverage and cost for individuals. Internal summarization, search, or drafting tools generally fall into lower tiers. Insurers should classify each use case explicitly and record the rationale, since tier drives the model risk duty set.
Do insurers deploying purchased AI have duties under the EU AI Act?
Yes. The Act distinguishes providers from deployers, and an insurer deploying a purchased high-risk system carries deployer obligations including human oversight, input data relevance, monitoring, and logging. Vendor sourcing never removes model risk work. Carrier inventories should record, for every system, whether the insurer acts as provider, deployer, or both, since the role determines which duties attach.
Does the EU AI Act affect insurers outside Europe?
It can. The Act's obligations extend extraterritorially to providers and deployers whose systems' output is used in the Union, regardless of where the organization is established. A North American carrier whose scoring or pricing models affect European customers may be in scope with no European entity. Map the insurance compliance footprint by data and decision flows rather than by corporate structure.