Short answer. Run one control framework, not two. DORA has applied since 17 January 2025, and the EU AI Act’s high-risk obligations for systems such as credit scoring were moved to 2 December 2027 by the Digital Omnibus. Both regimes demand the same five things from the same functions: board accountability, a risk management system, third-party vendor oversight, incident reporting, and documented evidence. A bank that maps the AI Act onto its live DORA framework reuses controls it already runs and avoids a second parallel program.
Most banks in the region already have a working DORA framework. The ICT risk register exists, the management body signs off on it, vendor contracts carry the required clauses, and the incident reporting chain is tested. When the EU AI Act lands on a high-risk system, the practical question is not “what new program do we build” but “which of our DORA controls already cover this, and where is the gap.” For a credit-scoring model, the answer is that most of the governance plumbing is already in place.
The work is to classify each AI system under Article 6 of the AI Act, then fold the extra AI-specific duties into the controls DORA already demands. One risk register gains an AI column. One incident timeline produces two notification packets when they are needed. One management body owns both. Ableneo has put 35 of its 49 AI projects into production and passed an ECB vendor security assessment, so the controls below are drawn from systems that actually run under supervision, not from a policy template.
The timing mismatch is the reason to act now rather than wait. DORA is already enforced across every EU financial entity. The AI Act’s high-risk deadline for Annex III systems like credit scoring moved to 2 December 2027, but classification under Article 6 applies regardless, and the Article 50 transparency duties for AI-generated content are already in effect. A bank that treats December 2027 as a distant problem will discover that the governance, documentation, and vendor controls it needs are the same ones DORA requires today.
For banks and insurers in Slovakia and the Czech Republic, the supervisory reality adds weight. The National Bank of Slovakia and the Czech National Bank examine ICT risk and third-party arrangements under DORA now. An AI system inside a regulated process will be read against both regimes in the same audit. Building one combined program, rather than two documents that contradict each other, is what keeps the audit short and the sign-off defensible.
Run one control framework: four of the five heaviest AI Act obligations map onto controls a DORA-compliant bank already operates.
They overlap on the parts that cost the most to build. Board accountability is the clearest case: DORA Article 5(2) makes the management body responsible for the ICT risk management framework, and the AI Act places deployer obligations under Article 26 at the same senior level. Third-party risk is the second: DORA Article 28 governs ICT vendor oversight and Article 30 sets the mandatory contract clauses, while the AI Act pushes provider documentation and conformity evidence through the same supplier relationship.
Risk management is the third overlap. DORA Article 6 requires a documented ICT risk management framework; AI Act Article 9 requires a risk management system for high-risk AI across its lifecycle. The same lifecycle controls (identify, protect, detect, respond, recover) carry both. The practical rule is that four of the five heaviest obligations map onto controls a DORA-compliant bank already runs.
The management body owns it in law under both regimes, so the day-to-day model has to roll up to that level cleanly. The pattern that holds in production is the three lines of defence already used for DORA: the business owns the AI system and its first-line controls, a second-line risk or compliance function owns the combined framework and the evidence, and internal audit reviews the whole thing. AI does not get its own parallel organisation. It gets a named owner inside the structure the bank already has, with the AI-specific expertise (model validation, bias testing) added to the second line rather than bolted on beside it.
An AI vendor can sit in the DORA register as an ICT third party and carry AI Act provider obligations at the same time. Handle it with one vendor file, not two reviews. Record the arrangement once in the DORA register of information, apply the Article 30 contractual provisions, and extend the same contract to cover the AI Act documentation the bank needs from a provider: the intended purpose, the risk classification, the data governance, and the information required for the bank’s own conformity evidence. The due-diligence questionnaire gains AI questions; it does not become a separate process.
Build the evidence base once. Both regimes are satisfied by the same underlying telemetry: logs that show the system behaved as documented, that humans could intervene, that incidents were detected and reported, and that the model was tested before and after changes. A bank that produces this from one monitoring layer answers a DORA resilience question and an AI Act oversight question from the same record. The single audit trail is the asset that makes a combined program cheaper to run than two, because the expensive part of compliance is producing defensible evidence, not writing policies.
Some duties belong to one regime only, and merging them is a mistake. The AI Act’s fundamental rights impact assessment, its conformity assessment, and its Article 50 transparency labelling of AI-generated content have no DORA equivalent. DORA’s threat-led penetration testing and its specific major-incident reporting windows have no AI Act equivalent. Treat these as distinct workstreams that sit inside the combined framework. One program, shared evidence, with a short list of regime-specific tasks that each have a clear owner and deadline.
A combined program only works if the AI systems underneath it actually reach production, because an audit judges what runs, not what was planned. Ableneo has put 35 of its 49 AI projects into production, passed an ECB vendor security assessment, and runs multi-year platforms for UNIQA across six markets and for Erste Group, all inside the same supervised environment DORA and the AI Act govern. That track record is why the controls described here come from live systems rather than a framework diagram. See the Ableneo services page for how this maps to a bank’s existing governance.
Key takeaways
Planning AI in a regulated business? Ableneo takes systems from classification to governed production.