Book a demo
Topics · AI-CONTROL-PLANE

Insurance AI Gateway DORA | Moring

DORA, the EU Digital Operational Resilience Act applicable since 17 January 2025, reaches AI systems in insurance by function rather than by name: any agent or model supporting a critical or important function falls inside its ICT risk framework. An insurance AI gateway operationalizes DORA for agents through identity, access control, logging, tested failover, and third-party governance at one enforcement point.

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

Short answer

DORA, the EU Digital Operational Resilience Act applicable since 17 January 2025, reaches AI systems in insurance by function rather than by name: any agent or model supporting a critical or important function falls inside its ICT risk framework. An insurance AI gateway operationalizes DORA for agents through identity, access control, logging, tested failover, and third-party governance at one enforcement point.

How it works

DORA is Regulation (EU) 2022/2554, applicable to in-scope financial entities including insurance and reinsurance undertakings since 17 January 2025. Its text does not reference AI, machine learning, or agents explicitly, and that is the point often missed: obligations are function-based, so any system, including an AI model or an agent, that supports a critical or important insurance function falls within scope because of what it does, not what it is called. AI in underwriting, claims triage, or automated decisioning cannot be treated as outside the perimeter.

The core obligation is the ICT risk management framework in Chapter II, Articles 5 through 16: a documented, board-approved framework covering identification, protection, detection, response, and recovery for ICT systems supporting critical or important functions. An agent adjudicating claims or a model suite supporting pricing is such a system, so the carrier must identify it, classify its criticality, protect it, detect anomalies, respond to incidents, and recover from disruption, with evidence at each step.

An AI gateway maps naturally onto those five capabilities for agent traffic. Identification: the gateway inventory records every agent and model path, with scoped identities replacing shared credentials. Protection: access policy, guardrails, and redaction bound what agents may reach. Detection: gateway telemetry surfaces anomalous usage and failed policy evaluations. Response and recovery: rate limits contain incidents, and a tested shutdown path stops a misbehaving agent while failover keeps critical service running.

Evidence quality determines whether the mapping satisfies supervisors. DORA compliance depends on timely, reliable records for incident investigation and reporting, and infrastructure practices cited for AI agents under DORA include append-only audit storage, per-agent identity attribution on every write, and enforced data residency. A gateway producing attributable, tamper-evident logs gives the carrier those properties for AI traffic as a default rather than a project.

Third-party governance is the other half. DORA extends oversight to ICT third-party service providers, and model providers are exactly that for an insurer renting frontier intelligence. The gateway centralizes which providers receive which traffic and records it, supporting the register and oversight expectations, while contractual and concentration-risk analysis proceeds through the carrier risk function. One control point seeing all provider traffic is materially easier to evidence than scattered integrations.

Positioning matters for the operating model. DORA obligations sit with the carrier, not with the gateway vendor, so the platform team should operate the gateway as the enforcement and evidence layer while compliance owns the framework documentation and board reporting. Carriers that integrate the two find that AI governance evidence and operational resilience evidence reinforce each other, which is precisely how supervisors now read AI use in insurance.

Frequently asked questions

Does DORA apply to AI agents used by insurance carriers?

Yes, by function rather than by name. DORA regulates ICT systems supporting critical or important functions of in-scope financial entities, including insurers, and it does not need to mention AI for an agent adjudicating claims or supporting pricing to be in scope. Since 17 January 2025, carriers must identify, protect, detect, respond, and recover for such systems under a board-approved framework, which makes the agent estate part of ICT risk management.

How does an AI gateway support the five DORA ICT risk capabilities?

The gateway identifies agent and model paths through its inventory and scoped identities, protects traffic with access policy, guardrails, and redaction, detects anomalies through telemetry and failed policy evaluations, and supports response and recovery through rate limits, tested failover, and a shutdown path for misbehaving agents. Each capability produces records, which is what supervisors examine. The gateway does not own the DORA framework; it enforces and evidences it for AI traffic.

What evidence does DORA expect for AI systems in insurance?

Reliable, timely records supporting incident investigation and reporting, produced under the ICT risk management framework in Chapter II. For AI agents specifically, cited infrastructure practices include append-only audit storage, per-agent identity attribution on every write, enforced data residency, and tested failover. A gateway that logs every model call and tool invocation with identities and policy decisions supplies these properties by default across claims and underwriting traffic.

How does DORA affect insurer relationships with model providers?

Model providers are ICT third-party service providers under DORA, so they enter the carrier third-party risk framework: registered, assessed, and overseen with attention to concentration risk. A gateway helps by centralizing which providers receive which traffic and recording every exchange, making the oversight register factual rather than aspirational. Contractual obligations remain a risk-function matter, but observable traffic is the evidence that makes those contracts meaningful.

Sources

  1. www.eiopa.europa.eu
  2. trussed.ai
  3. trustible.ai
  4. www.cockroachlabs.com
  5. www.furtherai.com
  6. aperion.ai
  7. www.cbh.com
  8. compassmsp.com
  9. - content.naic.org
  10. - artificialintelligenceact.eu
  11. - eur-lex.europa.eu