Book a demo
Topics · REGULATORY

Insurance AI Model Risk Management DORA | Moring

DORA, Regulation (EU) 2022/2554, applies to insurers and reinsurers authorized under Solvency II since 17 January 2025 and pulls AI systems into ICT risk management by function. Insurers should treat model risk controls, inventory, oversight, third-party registers, and resilience testing, as the DORA-facing surface.

On this page
  1. Short answer
  2. How it works
  3. FAQs
  4. Related reading
  5. Sources

Short answer

DORA, Regulation (EU) 2022/2554, applies to insurers and reinsurers authorized under Solvency II since 17 January 2025 and pulls AI systems into ICT risk management by function. Insurers should treat model risk controls, inventory, oversight, third-party registers, and resilience testing, as the DORA-facing surface.

How it works

DORA reaches insurers by function rather than by name. Regulation (EU) 2022/2554 became applicable to in-scope financial entities, including insurance and reinsurance undertakings authorized under Solvency II, from 17 January 2025. Its ICT risk management framework does not contain an AI chapter as such, but AI systems that support critical or important functions fall inside its scope by what they do. For carriers, that means model risk management and DORA compliance describe overlapping control surfaces over the same systems.

The practical consequence is that an underwriting model, a claims automation pipeline, or an agentic claims assistant is not merely a model risk subject; where it supports an important function, it is also an ICT-supported function with DORA obligations. Insurers therefore need to map which models sit inside critical or important functions, because that mapping determines which additional requirements attach: resilience testing, incident reporting, and register treatment for the supporting providers.

Third-party dependency is where DORA bites hardest on model risk. Many insurer AI models depend on external foundation models, inference providers, and data services, and DORA requires registers of information about ICT third-party providers with contractual provisions on access, audit, and exit. Model risk third-party assurance, audit rights, performance disclosure, change notification, and DORA's register and contract requirements reinforce each other and should be built once, with one evidence stream.

Incident management is the second hard seam. DORA requires classification and reporting of major ICT-related incidents, and model failures that disrupt critical functions, a claims pipeline stalling, an underwriting model producing systematically biased outputs, can qualify. Model risk escalation paths should therefore connect to DORA incident processes, with threshold definitions written jointly by the model risk function and operational resilience teams rather than each drafting its own.

Resilience testing completes the picture. DORA's testing expectations apply to ICT systems supporting critical functions, and model components within those systems should be included in scenario testing: what happens when the model is unavailable, degraded, or recalled. A tested rollback path for a claims model is simultaneously a model risk control and DORA resilience evidence, which is the kind of dual-purpose control that makes combined compliance efficient.

For insurers with European exposure, the efficient path is integration: one inventory that records both model risk tier and DORA functional classification, one third-party register feeding both vendor risk and DORA obligations, and one incident framework that routes model failures into ICT incident reporting where thresholds are met. Carriers that build these seams deliberately avoid running two governance programs over the same technology.

Frequently asked questions

Does DORA apply to AI models used by insurers?

Yes, by function. DORA, Regulation (EU) 2022/2554, has applied since 17 January 2025 to in-scope financial entities including insurance and reinsurance undertakings authorized under Solvency II. Its ICT risk framework has no AI chapter, but AI systems supporting critical or important functions fall inside scope by what they do, so a claims automation pipeline or agentic assistant carries DORA obligations alongside model risk controls.

How does DORA affect third-party AI dependencies for insurers?

DORA requires registers of information about ICT third-party providers with contractual provisions on access, audit, and exit. Insurer AI often depends on external foundation models, inference providers, and data services, so those relationships belong in the register. Model risk third-party assurance, audit rights, performance disclosure, and change notification, should be built once and feed both the model risk program and DORA obligations.

Can model failures trigger DORA incident reporting?

They can. DORA requires classification and reporting of major ICT-related incidents, and model failures that disrupt critical functions, a claims pipeline stalling or an underwriting model producing systematically biased outputs, can qualify. Model risk escalation paths should connect to DORA incident processes, with classification thresholds written jointly by model risk and operational resilience teams so nothing falls between them.

What is the efficient way to combine DORA and model risk compliance?

Integrate the seams deliberately: one inventory recording both model risk tier and DORA functional classification, one third-party register feeding vendor risk and DORA obligations, and one incident framework routing model failures into ICT reporting where thresholds are met. Rollback tests for critical models double as resilience evidence. Building these connections once avoids running two programs over the same systems.

Sources

  1. trussed.ai
  2. trustible.ai
  3. finalyse.com
  4. furtherai.com
  5. brainyyack.com
  6. cbh.com
  7. thecroforum.org
  8. optro.ai
  9. - content.naic.org
  10. - artificialintelligenceact.eu
  11. - eur-lex.europa.eu