DORA and AI Agents: Operational Resilience for Governed Workflows

DORA regulates ICT systems and digital operational resilience, not AI specifically. But an AI agent that acts inside a financial entity's production systems is part of the ICT estate: it accesses systems, changes records, and creates operational risk. Supervisors will expect the same logging, risk assessment, incident response and third-party controls for agent activity as for any other automated system. The governance burden lands whether or not the word "AI" appears in the regulation.

DORA requires financial entities to maintain detailed logs of events affecting ICT-supported functions, sufficient to reconstruct what happened. For agents, a defensible log captures: the agent identity (not a shared service account), every tool invocation and system change, the policy decision that allowed or refused the action, human approvals for gated actions, and timestamps consistent across systems. Logs an auditor can tamper with do not satisfy the requirement.

DORA's resilience testing applies to the functions agents support. A practical agent drill: trigger the stop control, verify no actions occur after the stop decision, produce a complete list of everything the agent wrote in a defined window, confirm each write has a rollback path or named human owner, and prove restart requires a named approver. If you cannot pass this drill, the agent is not operationally resilient, whatever the deployment dashboard says.

DORA applies to ICT third-party service providers that financial entities depend on, including cloud and critical ICT services. Model API providers are increasingly treated as part of the dependency chain. The practical consequence for you: an agent that calls an external model API creates a third-party dependency you must assess, contract for, and be able to exit — and the entity relying on the service carries the resilience obligation.

They are complementary. The EU AI Act regulates the AI system itself: classification, documentation, human oversight, conformity. DORA regulates the operational resilience of the systems the AI runs in: logging, incident response, testing, third parties. A bank deploying governed agents needs both: the AI Act asks "is this AI system governed?" while DORA asks "can this estate survive whatever it does?" Moring addresses both layers — agent governance for the AI Act, action-level evidence and stop controls for DORA.

Moring provides the controls DORA examiners will look for: per-agent identity so every action is attributable; immutable action records with policy decisions and human approvals; tested stop controls that halt new and in-flight actions with a complete write list; deployment inside your own cloud so the dependency stays yours to govern; and drill-ready evidence for resilience testing. Your agent workflow becomes an auditable ICT function rather than an ungoverned shadow system.