PrymeIntelligence

Regulation and standards·Definition

What is DORA?

DORA is the EU regulation on digital operational resilience for financial entities, applying from 17 January 2025. What it means for AI agents.

DORA, the Digital Operational Resilience Act (Regulation (EU) 2022/2554), is the EU law that requires banks, insurers, investment firms, payment firms and other financial entities to withstand, respond to and recover from ICT disruption. It has applied since 17 January 2025. AI agents are ICT systems, so DORA applies to them like any other.

Why does DORA matter for AI agents?

An AI agent in a financial firm is software, running on infrastructure, often supplied by third parties, connected to core systems and acting on them. Each of those facts brings it within DORA. If the agent supports a critical or important function, such as payments or claims handling, the expectations are higher.

DORA also changes how firms must think about agent vendors. A provider of an agent platform, or of the model behind it, is an ICT third-party service provider, with contract, oversight and exit requirements attached. That argument is made in DORA makes agent providers ICT third parties.

What does DORA require?

DORA has four main areas relevant to agents:

Area In substance For an agent
ICT risk management A framework to identify, protect, detect, respond and recover, owned by the management body Agents in the asset inventory, access controls, change control, a tested stop and rollback
ICT incident reporting Classify incidents and report major ones to the supervisor Agent failures assessed and reported like other ICT incidents
Digital operational resilience testing Regular testing of systems and recovery Agents included in resilience tests and scenario exercises
ICT third-party risk Contracts, register, oversight and exit plans for ICT suppliers Agent and model providers in the register, with exit plans

What does an example look like?

An EU payments firm deploys an agent that prepares outgoing payment batches. Under DORA, the firm records the agent and its model provider in its ICT asset inventory and third-party register, classifies payments as a critical function, and tests what happens if the model provider is unavailable. The answer, documented and tested, is that the agent drops to depth 1 and staff prepare batches manually under the firm's existing procedure. An agent malfunction that sends wrong payment files would be assessed against DORA's incident classification criteria. How DORA treats critical or important functions with AI agents explains why that classification drives so much else.

How is DORA different from the EU AI Act?

DORA is about resilience: keeping ICT systems running safely and recovering when they fail, whatever the technology. The EU AI Act is about the AI itself: risk classification, transparency, human oversight, accuracy. An agent used in an EU bank can fall under both. DORA asks whether the firm can withstand the agent or its provider failing; the AI Act asks whether the agent is designed and used in a way that protects people. The EU AI Act and DORA together for AI agents in banks shows how they fit.

It also differs from NIS2, which sets cybersecurity duties for essential and important entities more broadly; for financial entities, DORA takes precedence in its areas.

Who does DORA apply to?

EU financial entities across banking, insurance, investment, payments, crypto-asset services and others, plus critical ICT third-party providers designated for direct oversight. Whether DORA applies to AI agents in financial services answers the scope question in more detail. For the wider picture, see our guide to AI agent regulation.

This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

Questions people ask

Does DORA mention AI?

DORA is technology-neutral and does not single out AI. It applies to all ICT systems and services, so AI agents are covered by the same requirements as other software.

Is a model provider an ICT third-party service provider?

If it supplies an ICT service to a financial entity, such as model access over an API, it will generally be treated as one, with the contract and oversight requirements that follow.

Decide what your agents may do. Prove it. Enforce it.

Pryme Intelligence certifies each agent on your own cases, checks every action against five bounds at the moment it happens, and files every verdict.