Short answer. Four kinds of partner can help a Slovak bank adopt AI under DORA: Big 4 firms, global systems integrators, local AI consultancies and product vendors. DORA (Regulation (EU) 2022/2554, applicable since 17 January 2025) puts the accountability on the bank’s management body whichever one you pick, so the right choice is the partner whose contract, exit plan and delivery record you can defend in front of the supervisor. Choose on four tests: production track record, ICT third-party risk fit, exit plan, and concentration impact.
A Slovak bank that wants AI in production in 2026 hires a partner for one of three jobs: design the governance, build the system, or run it. Each job turns the partner into an ICT third-party service provider the moment it touches a production system. From that point DORA applies to the contract, not only to the project.
The market sorts into four groups. Each has a different strength and a different DORA cost.
DORA has applied to Slovak banks, insurers and investment firms since 17 January 2025. Article 28 requires a bank to manage ICT third-party risk as an integral part of its ICT risk framework, keep a register of information on every ICT contractual arrangement, and have exit strategies for services that support critical or important functions. The register format is set by Implementing Regulation (EU) 2024/2956, and the ESAs used the registers to designate the first critical providers.
An AI model API, a managed MLOps platform and a consultancy that operates your scoring pipeline all belong in that register. A partner chosen on a slide deck and papered later creates a finding. A partner chosen with the contract clauses in hand does not.
Four partner types serve a Slovak bank under DORA: Big 4, global SIs, local AI consultancies and vendors, and each one becomes an ICT third-party provider in your register.
The same five demands apply to every group, because Article 30 sets the minimum contract content. For services supporting critical or important functions, the contract must include full service descriptions, data location and protection terms, access, inspection and audit rights, incident assistance, participation in the bank’s resilience testing, and exit and transition provisions.
Article 29 requires a preliminary assessment of ICT concentration risk before a bank signs an arrangement for critical or important functions. The test is substitutability: could you move this service within a reasonable time, and how many of your critical services already sit with the same provider or its subcontractors?
For AI this bites on the model layer. If your AI service, your cloud and your core integration run on one provider, one outage or one contract dispute reaches all three. A bank that already has a hyperscaler as its main cloud should weigh a second AI provider, or an architecture where the model can be swapped, before adding a third dependency on the same group.
Article 28(8) requires exit strategies for ICT services that support critical or important functions. They must be documented, sufficiently tested and reviewed periodically. For AI, an exit plan names who holds the training data, prompts, evaluation sets and model configuration, in what format they come back, and how long a transition takes. A plan that says the vendor will assist fails that test.
Ask each candidate for a sample exit plan before you sign. A partner that cannot produce one has not run production AI for a regulated client.
The Národná banka Slovenska supervises Slovak banks and applies DORA through its supervisory process. For significant institutions, the ECB is the competent authority under the Single Supervisory Mechanism. In both cases the supervisor reads the register of information and the contracts, and asks the management body who accepted the risk. NBS has published DORA presentations since 2023 on its website. Confirm the current submission channel and deadlines for your register directly with your NBS supervisory team, since national practice differs between member states.
Ableneo is a local AI consultancy that builds and runs AI for banks, insurers and telcos in Slovakia, the Czech Republic and Austria. In 2025 Ableneo shipped 34 production AI projects, and roughly 4 of 5 projects reach production. One public example is T-Mobile Czech Republic, where document validation became 14x faster (7 hours to 30 minutes) at above 90% accuracy. Ableneo is not a designated critical provider and does not add to a bank’s concentration on the designated providers. Its strength is delivery depth in regulated environments. Read how the firm approaches regulated AI in the Ableneo AI Transformation FAQ, and test any partner, Ableneo included, against the checklist above.
Key takeaways
Planning AI in a regulated business? Ableneo takes systems from classification to governed production.