Insurance Agent Kill Switch DORA | Moring
DORA, the EU Digital Operational Resilience Act in force since January 2025, reaches insurance agents that act in production systems. For a carrier subject to it, the kill switch becomes a resilience control: containment within minutes, incident evidence on timeline, and tested recovery, demonstrated to an EU supervisor.
Short answer
DORA, the EU Digital Operational Resilience Act in force since January 2025, reaches insurance agents that act in production systems. For a carrier subject to it, the kill switch becomes a resilience control: containment within minutes, incident evidence on timeline, and tested recovery, demonstrated to an EU supervisor.
How it works
DORA changes the framing of the kill switch for carriers operating in the European Union. The regulation, in force since January 2025, applies to financial entities, insurers among them, and requires that the ICT systems they depend on be operationally resilient. An AI agent that changes a policy record, moves a payment, or touches claims infrastructure is part of that ICT estate, so the question DORA poses is not whether the agent is interesting technology but whether the carrier can keep operating safely when it malfunctions. Containment is the answer, and the kill switch is the mechanism.
Resilience under DORA is tested, not asserted. The regulation's philosophy, consistent with financial-sector operational resilience generally, is that capabilities must be demonstrated: failover exercised, incidents managed on defined timelines, evidence available for supervisory review. A kill switch satisfies this only in its drilled form. Runtime containment guidance describes minutes-level halting as depending on three preconditions, distinct agent identities so one agent can be stopped without collateral damage, controls enforced outside the agent's own logic, and graduated stop tiers. Each precondition is a DORA-relevant capability, and each should appear in the carrier's resilience testing plan.
Incident reporting sharpens the evidence requirement. DORA contemplates timely, reliable reporting of major ICT incidents, and containment records are central evidence: what the agent did, when the stop was invoked, by whom, which mechanisms applied, and the verified outcome of each layer. Kill switch guidance emphasizes preserving a complete record of agent actions before termination, and under DORA that record feeds the incident timeline the supervisor will examine. Carriers should align their kill switch invocation record format with their DORA incident-reporting templates before an incident, not during one.
DORA also reaches third parties, which matters because many carriers buy agent capability from vendors. If a vendor platform hosts or executes an insurance agent, the carrier remains responsible for ensuring resilience, and contractual arrangements must secure the containment path: the carrier's operators must be able to halt the vendor-hosted agent without waiting on a support ticket. Kill switch architecture decisions, where the control plane lives, who holds credentials, how evidence is exported, should be negotiated in contracting rather than discovered during an incident.
The synthesis for an EU-scope carrier is to treat the kill switch as part of the resilience framework that DORA makes examinable: mapped into the ICT asset inventory, covered in threat-led testing where proportionate, wired into incident procedures with named roles, and evidenced in invocation records. Outside the European Union, the same design holds up under NAIC expectations for governed AI use, and the EU AI Act's high-risk obligations for underwriting and pricing agents add the oversight dimension. One containment architecture, documented once, answers all three.
Frequently asked questions
Does DORA apply to AI agents at an insurance carrier?
Yes, where the agent acts within the carrier's ICT estate. DORA, in force since January 2025, applies to financial entities in the European Union and requires resilience of the ICT systems they depend on. An agent that changes policy records, moves payments, or touches claims infrastructure is inside that scope regardless of where its compute runs. Containment capability, minutes-level halting with tested procedures, becomes a resilience control the supervisor can expect to see demonstrated.
What DORA-aligned properties should a kill switch have?
Tested, tiered, and evidenced. Testing aligns with DORA's philosophy that resilience is demonstrated through exercised scenarios, so containment drills belong in the resilience testing plan. Tiering, distinct agent identities, controls outside the agent's own logic, and graduated stop mechanisms, enables scoped containment without collateral damage. Evidence, invocation records showing trigger, invoker, mechanisms, and verified outcome per layer, feeds the timely, reliable incident reporting DORA contemplates for major ICT incidents.
How does DORA affect vendor-hosted insurance agents?
The carrier remains responsible for operational resilience even when the agent platform is third-party. Contracting must secure the containment path: carrier operators able to halt a vendor-hosted agent without a support-ticket delay, clarity on who holds control-plane credentials, and exportable evidence of invocations and agent actions. Architecture decisions about where the control plane lives should be negotiated in the contract, because discovering them during an incident leaves the carrier unable to meet its own DORA obligations.
How should containment records align with DORA reporting?
Before an incident, not during one. The invocation record format should map to the carrier's incident-reporting templates: agent actions before termination, time of stop, invoker identity and role, mechanisms applied, and verified outcome per layer. Kill switch guidance stresses preserving a complete record of what the agent did, which under DORA becomes the timeline backbone of the report. Rehearsing the mapping in a tabletop exercise keeps the alignment real under pressure.