Payments Model Risk EU AI Act | Moring
The EU AI Act, Regulation (EU) 2024/1689 published on 12 June 2024, sorts AI systems into risk tiers and attaches escalating obligations to each, with credit scoring and related financial use cases named directly as high risk. Its scope is extraterritorial, reaching payment institutions outside the Union whose model outputs are used there, and its phased milestones continue through 2026 and beyond.
Short answer
The EU AI Act, Regulation (EU) 2024/1689 published on 12 June 2024, sorts AI systems into risk tiers and attaches escalating obligations to each, with credit scoring and related financial use cases named directly as high risk. Its scope is extraterritorial, reaching payment institutions outside the Union whose model outputs are used there, and its phased milestones continue through 2026 and beyond.
How it works
The EU AI Act is Regulation (EU) 2024/1689, laying down harmonised rules on artificial intelligence, published on 12 June 2024, and it is the binding European framework that payment institutions with Union exposure must plan against. The Act moves regulation toward phased implementation in 2026, and it changes the governance of AI systems fundamentally: AI is no longer treated as a neutral technology but is evaluated on the level of risk it poses to security, fundamental rights, and social impact.
Sources: thebanker.com: https://www.thebanker.com/provision/ecdf1ee0-9301-5031-8031-700ac400f3cb adherent.com: https://www.adherent.com/blog/ai-governance-under-the-eu-ai-act-risk-classification-and-compliance-readiness-for-2026/
The Act's method is risk tiers with escalating obligations, from light-touch transparency rules up to strict pre-market conformity requirements for systems classified as high risk. Financial services carries the highest concentration of explicitly named high-risk use cases of any sector the Act covers: credit scoring, SME lending, probability-of-default and loss-given-default models, and insurance underwriting are all named directly in Annex III. That makes the tiers a direct compliance question for payment and banking groups, not a background policy debate.
Sources: linkedin.com: https://www.linkedin.com/pulse/what-eu-ai-acts-risk-based-approach-means-banks-regulators-outside-4y94f
Scope is deliberately extraterritorial. Article 2 applies the Act to providers and deployers established outside the European Union whenever the output of their AI system is used in the Union, regardless of where the institution or its model vendor sits. For payments groups this means a fraud model, a scoring service, or an agent operated from another region falls in scope if its decisions touch Union customers, and the compliance obligation follows the output, not the headquarters.
Sources: linkedin.com: https://www.linkedin.com/pulse/what-eu-ai-acts-risk-based-approach-means-banks-regulators-outside-4y94f
The Act sits alongside DORA, and the two must not be confused. They are separate regulations of equal legal status operating on different objects: DORA addresses ICT risk management for in-scope financial entities, including payment institutions, while the EU AI Act addresses AI systems placed on the market, put into service, or used in the Union. A payment institution using AI manages ICT risk under DORA and applies AI Act obligations separately where its systems and role, provider or deployer of high-risk AI, fall within the Act's scope.
Sources: docs.modulos.ai: https://docs.modulos.ai/frameworks/comparison/eu-ai-act-vs-dora
Compliance readiness for 2026 turns on documentation and monitoring. The Act's phased implementation shifts companies toward structured governance of AI systems, their documentation, and their monitoring, so payment institutions should identify where AI is used, classify systems by risk, and build the evidence base, technical documentation, human oversight arrangements, and logging, that high-risk classification demands. The third-party dimension matters too, because vendor-operated models serving Union payments make the institution a deployer with deployer obligations.
Sources: adherent.com: https://www.adherent.com/blog/ai-governance-under-the-eu-ai-act-risk-classification-and-compliance-readiness-for-2026/ empoweredsystems.com: https://empoweredsystems.com/blog/navigating-new-frontiers-understanding-and-adapting-to-the-eus-ai-act-for-enhanced-third-party-risk-management/
The compliance milestone of 2 August 2026 concentrates the work. Financial institutions should use the period to map their model estate against the Act's categories, decide provider and deployer roles per system, and align AI management with existing operational resilience requirements so evidence is produced once and used by both regimes. Institutions that wait for the milestone date will find classification, documentation, and oversight work that cannot be compressed into the weeks before it.
Sources: adherent.com: https://www.adherent.com/blog/ai-governance-under-the-eu-ai-act-risk-classification-and-compliance-readiness-for-2026/
The practical programme for a payments group is to inventory models with Union touchpoints, classify each against the Act's tiers with the role assignment recorded, close documentation and human oversight gaps for anything high risk, and contract vendors for the evidence deployers must hold. Extraterritorial scope means this work belongs to the group risk function rather than to a European subsidiary alone, because the output of a model is what places it in scope.
Sources: linkedin.com: https://www.linkedin.com/pulse/what-eu-ai-acts-risk-based-approach-means-banks-regulators-outside-4y94f docs.modulos.ai: https://docs.modulos.ai/frameworks/comparison/eu-ai-act-vs-dora
Third-party risk management is where much of the readiness work lands. The Act's arrival has reshaped how institutions approach vendors whose models serve their customers, and payment groups should respond by extending due diligence into the AI layer: which vendor systems classify as high risk, what documentation the deployer role requires, and whether contracts deliver the evidence obligations the Act attaches. A vendor model serving Union payment decisions makes the institution accountable for oversight it must be able to exercise, and exercising that oversight begins with knowing the classification of everything the vendor operates on the institution's behalf, then holding the vendor to it contractually.
Frequently asked questions
Which payment and financial use cases does the EU AI Act treat as high risk?
The Act names financial services use cases directly in Annex III, including credit scoring, SME lending, probability-of-default and loss-given-default models, and insurance underwriting, giving the sector the highest concentration of explicitly named high-risk use cases of any sector covered. Payment-adjacent systems that carry similar decision weight should be assessed against the same logic. High-risk classification attaches escalating obligations, from strict pre-market conformity requirements to documentation, human oversight, and monitoring duties.
Does the EU AI Act apply to a payment institution outside the European Union?
Yes, by design. Article 2 sets extraterritorial scope: the Act applies to providers and deployers established outside the Union whenever the output of their AI system is used in the Union. A fraud model, scoring service, or agent operated from another region is in scope if its decisions touch Union customers. The obligation follows the model output rather than the headquarters location, which places the work with the group risk function.
How do the EU AI Act and DORA differ for payment institutions?
They are separate regulations of equal legal status operating on different objects: DORA addresses ICT risk management for in-scope financial entities, while the EU AI Act addresses AI systems placed on the market, put into service, or used in the Union, with obligations depending on the system's classification and the entity's role as provider or deployer. A payment institution using AI manages ICT risk under DORA and applies AI Act duties separately where its systems fall within scope, so one integrated programme over a single inventory is the efficient structure.
What should payment institutions do before the 2026 implementation milestones?
Inventory the model estate with Union touchpoints, classify each system against the Act's risk tiers, record the provider or deployer role per system, and close documentation, human oversight, and monitoring gaps for anything classified high risk. The phased implementation shifts obligations toward structured documentation and monitoring, and aligning the work with existing operational resilience programmes lets evidence serve both regimes. Vendor-operated models require contractual evidence rights, since using them makes the institution a deployer.