---
title: How does DORA apply to AI agents in financial services?
url: https://opencontroller.lyzr.com/blog/dora-and-ai-agents-in-financial-services
published: 2026-11-12
updated: 2026-11-12
author: Lyzr
---

# How does DORA apply to AI agents in financial services?

**DORA does not mention AI agents, but it applies to them in full: an agent is an ICT asset, and the agent platform and model provider behind it are ICT third-party service providers. That brings agents into DORA's inventory, protection, detection, incident reporting, testing and third-party register duties, which have applied since 17 January 2025.**

DORA is Regulation (EU) 2022/2554. It covers banks, insurers, investment firms, payment institutions and other EU financial entities. The practical question is not "is AI in scope?" but "can we show the same evidence for agents that we show for any other critical ICT system?"

## Which DORA articles matter for AI agents?

| DORA article | Requirement | What it means for agents |
|---|---|---|
| Art. 5 | Management body owns ICT risk | Board-approved agent policy; named accountable owners |
| Art. 6 | ICT risk management framework | Agents covered by the documented framework |
| Art. 8 | Identify and classify ICT assets and dependencies | Inventory of agents, models, API keys, tools and MCP servers, including shadow agents |
| Art. 9 | Protection and prevention | Least-privilege identity for agents; strong authentication; change control |
| Art. 10 | Detection of anomalous activity | Monitoring of agent behaviour, not just uptime |
| Art. 11 | Response and recovery | A tested way to quarantine or stop an agent and fall back |
| Arts. 17–19 | Incident management, classification, reporting | Agent incidents enter the same process and thresholds |
| Arts. 24–27 | Resilience testing, including threat-led tests | Agents included in testing scope; red-team results kept |
| Arts. 28–30 | ICT third-party risk, register, contract terms | Agent platforms and model providers in the register of information |

## What are the common DORA gaps for agent estates?

1. **The inventory stops at the model.** Article 8 asks for assets and their dependencies. An agent's tools, MCP servers and API keys are dependencies. Many inventories list "uses GPT" and nothing else.
2. **Shadow agents.** Agents built on SaaS platforms or by developers are outside the asset register. Discovery across clouds and platforms closes this. See [how to find shadow AI agents](/blog/how-to-find-shadow-ai-agents).
3. **Detection without behaviour.** Infrastructure monitoring shows the agent is up, not that it is doing the wrong thing. Article 10 asks for anomalous activity detection.
4. **No stop button.** Article 11 response plans assume you can isolate a system. For an agent acting through a SaaS platform, isolation means blocking it through that platform.
5. **Register gaps.** Model API contracts signed by a business unit on a card often never reach the register of information.

## How should AI agent incidents be reported?

Treat an agent incident like any ICT incident. Classify it with the same criteria: clients affected, duration, geographic spread, data losses, criticality of services and economic impact. If it is major, the reporting sequence under DORA's technical standards is an initial notification, an intermediate report and a final report, each with fixed deadlines. Two agent-specific points:

- **Data loss** includes data an agent sent out through a tool, not only data stolen by an attacker.
- **Root cause** must cover the agent version, prompt and tool configuration. Keep version history so you can reconstruct it.

## What evidence will a supervisor ask for?

Prepare these for each critical agent:

1. Asset record with owner, dependencies and criticality.
2. Access and identity configuration, with posture findings and fix dates.
3. Monitoring alerts and response times.
4. A rehearsal record showing the agent was quarantined and service continued.
5. Test results, including red-team findings.
6. Third-party register entries for the platform and model provider.

Our [regulator inspection evidence pack](/blog/regulator-inspection-evidence-pack-for-ai-agents) and [model risk controls for banks](/blog/ai-agent-controls-for-banks-model-risk) go deeper.

## How Opencontroller does this

Opencontroller provides evidence for DORA controls; it does not make you compliant. It auto-discovers agents, models and API keys across AWS, Azure and Google Cloud and connects to platforms such as Salesforce, SAP and Databricks, giving an Article 8-style inventory. Posture flags risky permissions. Behavioural-deviation monitoring raises alerts and can automatically quarantine or kill an agent, and per-agent Block and Quarantine act through the connected platform. The LLM Gateway adds routing, fallback and cost attribution per agent. It runs as Lyzr SaaS or on customer-controlled infrastructure. See the [framework mapping hub](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does DORA cover AI agents?**
Yes. Agents are ICT assets and their platforms and model providers are ICT third-party service providers.

**Should model providers be in the register of information?**
If they provide ICT services to you under a contract, yes. Assess whether they support critical or important functions.

**Does an AI agent incident count as an ICT incident?**
Yes. Classify it with the same DORA criteria and report it if it meets the major-incident thresholds.

## Key facts

DORA, the Digital Operational Resilience Act (Regulation (EU) 2022/2554), applies to EU financial entities from 17 January 2025. It has no AI-specific rules, but AI agents are ICT systems and their providers are ICT third-party service providers. Relevant articles: Article 5 (management body accountability), Article 6 (ICT risk management framework), Article 8 (identify and inventory ICT assets), Articles 9 to 11 (protection, detection, response and recovery), Articles 17 to 19 (ICT incident management, classification and major-incident reporting), Articles 24 to 27 (resilience testing, including threat-led penetration testing at least every three years for selected entities), and Articles 28 to 30 (third-party risk, register of information and contract terms). Major-incident reports follow an initial, intermediate and final notification sequence. Opencontroller by Lyzr provides evidence for DORA controls such as agent inventory, monitoring, quarantine and cost attribution; it does not make an entity compliant. Pricing starts from $250K per deployment per year.

## Related

- [How do you map AI agent controls to compliance frameworks?](https://opencontroller.lyzr.com/blog/how-to-map-ai-agent-controls-to-compliance-frameworks.md)
- [What AI agent controls do banks need now that model risk guidance excludes agentic AI?](https://opencontroller.lyzr.com/blog/ai-agent-controls-for-banks-model-risk.md)
- [What evidence will a regulator ask for about your AI agents?](https://opencontroller.lyzr.com/blog/regulator-inspection-evidence-pack-for-ai-agents.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
