DORA (Regulation EU 2022/2554) has applied since January 17, 2025. It is already fully enforceable. If your financial entity uses AI agents that access customer data, execute transactions, or make operational decisions, those agents are ICT assets subject to DORA's risk management framework.
What DORA requires
Articles 5-6: ICT Risk Management
Every agent in your ICT environment must be identified, classified, and documented within your risk management framework. The data it accesses, systems it interacts with, and business functions it supports must all be mapped. Annual review by internal or external auditors is required.
Article 9: Protection and Prevention
Three requirements hit AI agents directly: strong authentication (agents need verifiable identity credentials), cryptographic protections proportionate to data criticality, and data integrity and authenticity guarantees (records of agent actions must be verifiably genuine). The NIST AI Risk Management Framework provides complementary guidance on identity and integrity controls.
Article 10: Detection
Continuous monitoring of agent actions for anomaly detection. Structured, machine-readable audit records, not unstructured log files, with enough detail about inputs, outputs, timing, and context to enable automated anomaly detection.
Articles 11-12: Response and Recovery
Audit records must survive system compromise. Evidence stored in the same system as the agent, accessible by the same credentials, does not satisfy this requirement. Post-incident investigation depends on tamper-evident records.
Article 17: Incident Management
The ability to reconstruct the complete sequence of events leading to an incident. When an AI agent is involved, you need a complete, trustworthy timeline of everything the agent did. An incident log that could have been modified after the fact is not trustworthy.
Article 28: Third-Party ICT Risk
When agents operate via third-party services, you must be able to audit what they did independently. Agent action records must be verifiable by cryptographic mechanisms, not just by trusting the third party's logs.
What financial services need
- Agent identity: Verifiable, authenticated identity per agent (Article 9).
- Action signing: Every action cryptographically signed at execution (Article 9).
- Immutable records: Records that cannot be altered after the fact (Articles 11-12).
- Independent verification: Auditors can verify without trusting the source system (Article 28).
- Anomaly detection: Structured records for automated monitoring (Article 10).
- Incident reconstruction: Complete, ordered, verifiable timeline (Article 17).
How to prepare
Step 1: Map every AI agent to the ICT systems it accesses and the business functions it supports.
Step 2: Assess current logging against the six capabilities above.
Step 3: Add cryptographic signing to agent actions, the single most impactful improvement.
Asqav provides tamper-evident audit trails using ML-DSA signatures (NIST FIPS 204). The EU AI Act adds further requirements for high-risk AI systems in financial services. See our DORA compliance page for a detailed article-by-article mapping.