Eighteen markets. One EU rulebook that moved, and a long tail of supervisors still asking their own questions.

Oonagh van den Berg | RAW Compliance

Issue 3 mapped the Middle East, where one model met several different gates. Europe looks more uniform from a distance. There is one AI Act. Up close, it is a layered map: the AI Act sets the horizontal rule, DORA and the banking rulebook already apply, and national supervisors add their own expectations about governance, explainability and third parties.

The headline for bank boards is a date that moved. The EU has now postponed the high-risk obligations that matter most to banks. That is a timing change, not a reason to stop building controls.

European Union - the AI Act, with a later date for high-risk

The AI Act entered into force on 1 August 2024. The Digital Omnibus on AI changed part of its timetable. The Council and Parliament reached a provisional agreement on 7 May 2026, Parliament gave final approval on 16 June and the Council adopted the text on 29 June. The amending text is Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026 and in force. It delays the high-risk obligations only. Standalone high-risk systems, which include creditworthiness and credit scoring of natural persons, now apply from 2 December 2027 instead of 2 August 2026. High-risk systems embedded in regulated products apply from 2 August 2028. The Council describes the delay as applying to the high-risk rules, not to the whole Act. The Council also lists national AI sandboxes due by 2 August 2027, a grace period for marking AI-generated content that ends on 2 December 2026, and a new prohibition on AI that generates non-consensual sexual content or child sexual abuse material.

The European Banking Authority’s banking-sector paper, updated in November 2025, makes two points that still stand. Creditworthiness assessment of natural persons is high-risk under the Act. And CRD, CRR, DORA, the consumer and mortgage credit directives, the payment services rules and EBA guidelines already apply to banks using AI. Delay does not remove those duties.

Sources: https://eur-lex.europa.eu/eli/reg/2026/1744/oj/eng ; https://www.consilium.europa.eu/en/press/press-releases/2026/06/29/artificial-intelligence-council-gives-final-green-light-to-simplify-and-streamline-rules/ ; https://www.consilium.europa.eu/en/press/press-releases/2026/05/07/artificial-intelligence-council-and-parliament-agree-to-simplify-and-streamline-rules/ ; https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act ; https://eba.europa.eu/sites/default/files/2025-11/d8b999ce-a1d9-4964-9606-971bbc2aaf89/AI%20Act%20implications%20for%20the%20EU%20banking%20sector.pdf

United Kingdom - no AI Act, a live testing lane and a model-risk rulebook

The UK has no AI Act. Banks answer to existing duties. The PRA’s supervisory statement SS1/23 sets model risk management principles for banks, and a firm using AI in a model-driven decision should read it as the starting point. The FCA runs AI Live Testing, a supervised route for firms to test AI applications. In April 2026 it named a second cohort of eight firms, including Barclays, Experian, Lloyds Banking Group (Scottish Widows) and UBS. On 8 October 2026 the ICO published a report saying ten foundation model developers have made or committed to data protection changes, and opened a call for evidence on agentic AI. A live testing slot is a supervised test, not an approval to go live.

Sources: https://www.bankofengland.co.uk/prudential-regulation/publication/2023/may/model-risk-management-principles-for-banks-ss ; https://www.fca.org.uk/news/press-releases/fca-announces-second-cohort-ai-live-testing ; https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2026/10/ico-secures-changes-from-leading-ai-developers-as-scrutiny-extends-to-ai-agents/

Germany - BaFin ties AI to DORA

BaFin’s guidance on ICT risks in the use of AI at financial entities is dated 23 January 2026. It is non-mandatory. It helps banks and insurers apply DORA when they use AI, covering ICT risk management and ICT third-party risk, including subcontracting. The practical reading: an AI system is an ICT asset and an ICT dependency. If a vendor model sits in a critical process, the DORA register, exit plan and subcontracting chain apply to it.

Source: https://www.bafin.de/SharedDocs/Downloads/EN/Anlage/dl_Anlage_orientierungshilfe_IKT_Risiken_bei_KI_en.pdf?__blob=publicationFile&v=3

France - the ACPR reference on governance and explainability

The ACPR’s reflection paper on the governance of AI algorithms in the financial sector remains the French reference. It grew out of exploratory work with financial firms that the ACPR launched in March 2019. Its focus is explainability and governance of machine-learning systems. It is a supervisory discussion document, not a binding rule. The ACPR has since published a report on the legal and regulatory impacts of AI in banking, finance and insurance (2025). Expect questions about who validated the model and how its output can be explained to a customer or a supervisor.

Sources: https://acpr.banque-france.fr/fr/publications-et-statistiques/publications/gouvernance-des-algorithmes-dintelligence-artificielle-dans-le-secteur-financier ; https://acpr.banque-france.fr/fr/system/files/2025-07/Rapport_68_F_V3.pdf

Netherlands - the SAFEST principles

De Nederlandsche Bank’s general principles for the use of AI in the financial sector use the acronym SAFEST: soundness, accountability, fairness, ethics, skills and transparency. DNB and the AFM have also published a joint report on the impact of AI on the financial sector and supervision, which says AI use is part of the existing processes and services they already supervise. These are supervisory principles and a supervisory report, not legislation. They give a Dutch bank a clear checklist: who is accountable, what has been tested for soundness and fairness, and whether staff have the skills to challenge the model.

Sources: https://www.dnb.nl/media/jkbip2jc/general-principles-for-the-use-of-artificial-intelligence-in-the-financial-sector.pdf ; https://www.dnb.nl/media/gepbbikm/ai.pdf

Ireland - AI as a supervisory spotlight

The Central Bank of Ireland’s 2026 Regulatory and Supervisory Outlook includes a spotlight on approaching AI from a supervisory perspective and an appendix on the AI landscape. It sets the tone: firms are expected to keep strong governance, risk management and operational resilience, and to act in customers’ best interests. It names bias, misinformation, data loss and “black box” opacity as risks. It is an outlook, not a rulebook, but it tells Irish firms where supervisors will look.

Source: https://www.centralbank.ie/docs/default-source/publications/regulatory-and-supervisory-outlook-reports/regulatory-supervisory-outlook-report-2026.pdf

Switzerland - FINMA guidance on AI governance

Switzerland is outside the EU, so the AI Act does not apply directly. FINMA published Guidance 08/2024 on governance and risk management when using AI on 18 December 2024. It describes what FINMA has seen in supervision. It names model risks such as lack of robustness, correctness, explainability and bias, data risks, IT and cyber risks, third-party dependencies, and legal and reputational risks. FINMA observed that most institutions were still early in building their governance. A Swiss bank selling into the EU still has to meet the AI Act for those customers.

Source: https://www.finma.ch/en/news/2024/12/20241218-mm-finma-am-08-24/

Luxembourg - a documentation set from the CSSF

The CSSF has built a set of supporting documents on AI for regulated entities: a white paper on opportunities, risks and recommendations (2018), questions and answers (1 August 2024) and and a 2024 thematic report on AI, all listed on the CSSF’s AI page. This is guidance and observation, not a separate approval regime. Read the thematic report for what the CSSF observed.

Sources: https://www.cssf.lu/en/artificial-intelligence/

Italy - a national AI law that names the financial supervisors

Italy passed its own AI law, Law 132 of 23 September 2025, in force from 10 October 2025. The text keeps the Bank of Italy, Consob and IVASS as the market surveillance authorities for AI systems used in their sectors, so Italian financial AI sits under supervisors the bank already knows. The Bank of Italy has also published a report with the OECD on AI in Italian financial markets (April 2026). Read the law with the EU AI Act, not instead of it.

Sources: https://www.normattiva.it/eli/id/2025/09/25/25G00143/CONSOLIDATED ; https://www.bancaditalia.it/media/notizie/2026/AI-in-Italian-Financial-Markets-en.pdf?language_id=1

Spain - a dedicated AI supervisory agency

Spain has a standalone AI supervisor, the Spanish Agency for the Supervision of Artificial Intelligence (AESIA). For financial firms, the Governor of the Banco de España has said the AI Act gives the national authority responsible for financial supervision the market surveillance role for high-risk financial models. A Spanish bank should expect to deal with AESIA on general AI Act questions and with the Banco de España on financial models.

Sources: https://aesia.digital.gob.es/ ; https://www.bde.es/f/webbe/GAP/Secciones/SalaPrensa/IntervencionesPublicas/Gobernador/Arc/Fic/IIPP-2024-10-07-escriva-en-tr.pdf

The Nordics - supervisors lead with surveys, recommendations and sandboxes

Sweden: Finansinspektionen reported in December 2024 that AI use is spreading quickly in the financial sector while risk management is lagging behind. Denmark: the Danish FSA published recommendations on supervised machine learning in 2019 and an updated set on AI in 2024. Norway: Finanstilsynet ran a regulatory sandbox project on AI in a digital pension adviser and warned in March 2025 against using AI for investment decisions. Finland: FIN-FSA published a thematic review of AI use in the financial sector on 27 June 2025, and a government proposal on AI Act supervision went to Parliament in May 2025. None of these is a binding AI rule. All show what the supervisor will ask.

Sources: https://www.fi.se/en/published/reports/reports/2024/ai-increasingly-common-in-the-financial-sector-but-risk-management-is-lagging-behind/ ; https://www.dfsa.dk/financial-themes/fintech/guidance-and-handheld-supervision/recommendations-when-using-artificial-intelligence ; https://www.finanstilsynet.no/nyhetsarkiv/nyheter/2025/finanstilsynet-advarer-mot-a-bruke-kunstig-intelligens-til-finansielle-investeringer/ ; https://www.finanssivalvonta.fi/en/publications-and-press-releases/supervision-releases/2025/thematic-review-of-the-use-of-AI-in-the-financi

Austria - existing technology-neutral rules, DORA and the AI Act

The FMA says supervision of AI use is already within its remit. AI used for regulated services must meet existing governance, conduct and distribution rules and DORA, and the management body keeps full responsibility for ICT risk, including risk from AI use and from AI-assisted cyber attacks. The AI Act adds to that. There is no separate Austrian AI approval step.

Source: https://www.fma.gv.at/en/cross-sectoral-topics/artificial-intelligence/

Poland - cyber resilience recommendations on advanced AI

Reported, not regulator-confirmed: In late July 2026 the UKNF, the KNF’s supervisory office, issued recommendations to financial market firms on advanced AI models. They say AI can change the cyber threat landscape and raise pressure on digital operational resilience. This comes from CyberDefence24 coverage and has not been checked against the UKNF’s own page, so treat the detail as secondary until confirmed.

Source: https://cyberdefence24.pl/cyberbezpieczenstwo/uknf-sektor-finansowy-musi-byc-gotowy-na-zaawansowana-ai

Belgium and Portugal - no AI-specific bank instrument found

The sources checked for this issue show no AI-specific bank rule from the National Bank of Belgium, the FSMA, the Banco de Portugal or Portuguese supervisors. The NBB runs a fintech and innovation page and the Banco de Portugal runs a Fintech+ contact channel. That is a scoped finding, not proof that nothing exists. Both banks still carry the AI Act, DORA and existing governance duties.

Sources: https://www.nbb.be/en/financial-supervision/fintech-insurtech-innovation ; https://www.bportugal.pt/page/fintech?mlid=2852

Why this matters beyond Europe

Europe shows what happens when a binding horizontal law meets existing sector rules. The law is not the whole map. A later date for high-risk AI in the EU does not delay DORA, model-risk expectations, consumer-credit duties or supervisory questions on governance. Compare it with the earlier issues. Kenya proposed an approval gate. Qatar has two approval triggers. APAC is mostly consultation and guidance. Europe has a fixed legislative timetable and a long supervisory tail. Each answer to “who approved this for production?” is different.

What bank boards should do now

1. Keep building the AI inventory and the high-risk classification, especially for credit scoring. The 2 December 2027 date is a deadline for the Act’s high-risk duties, not a pause on supervisory scrutiny.

2. Map each AI system to DORA: register entry, critical function, vendor, subcontractors and exit plan.

3. Record who validated each model, who challenged it and who can switch it off. This is the common question from the PRA, the ACPR, DNB and FINMA.

4. Treat a sandbox or live-testing slot as a test, not as approval for production.

5. Watch 2 December 2026 for AI-generated content marking and 2 August 2027 for national sandboxes.

Next issue: Latin America.

Source and status note

Positions checked against regulator and institutional materials available on 10 and 11 October 2026. The Digital Omnibus is cited from EUR-Lex (Official Journal, 24 July 2026, in force) and the Council’s final adoption release. Poland is from press coverage and is not yet confirmed from the UKNF. Belgium and Portugal are scoped findings. The CSSF page was last updated in 2024. Several national documents are older supervisory papers and may have been superseded. This is a market map, not legal advice.

THE AI REGULATION MARKET MAP • EUROPE • RAW COMPLIANCE

Eighteen markets. One EU rulebook that moved, and a long tail of supervisors still asking their own questions.

Oonagh van den Berg | RAW Compliance

Issue 3 mapped the Middle East, where one model met several different gates. Europe looks more uniform from a distance. There is one AI Act. Up close, it is a layered map: the AI Act sets the horizontal rule, DORA and the banking rulebook already apply, and national supervisors add their own expectations about governance, explainability and third parties.

The headline for bank boards is a date that moved. The EU has now postponed the high-risk obligations that matter most to banks. That is a timing change, not a reason to stop building controls.

European Union - the AI Act, with a later date for high-risk

The AI Act entered into force on 1 August 2024. The Digital Omnibus on AI changed part of its timetable. The Council and Parliament reached a provisional agreement on 7 May 2026, Parliament gave final approval on 16 June and the Council adopted the text on 29 June. The amending text is Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026 and in force. It delays the high-risk obligations only. Standalone high-risk systems, which include creditworthiness and credit scoring of natural persons, now apply from 2 December 2027 instead of 2 August 2026. High-risk systems embedded in regulated products apply from 2 August 2028. The Council describes the delay as applying to the high-risk rules, not to the whole Act. The Council also lists national AI sandboxes due by 2 August 2027, a grace period for marking AI-generated content that ends on 2 December 2026, and a new prohibition on AI that generates non-consensual sexual content or child sexual abuse material.

The European Banking Authority’s banking-sector paper, updated in November 2025, makes two points that still stand. Creditworthiness assessment of natural persons is high-risk under the Act. And CRD, CRR, DORA, the consumer and mortgage credit directives, the payment services rules and EBA guidelines already apply to banks using AI. Delay does not remove those duties.

Sources: https://eur-lex.europa.eu/eli/reg/2026/1744/oj/eng ; https://www.consilium.europa.eu/en/press/press-releases/2026/06/29/artificial-intelligence-council-gives-final-green-light-to-simplify-and-streamline-rules/ ; https://www.consilium.europa.eu/en/press/press-releases/2026/05/07/artificial-intelligence-council-and-parliament-agree-to-simplify-and-streamline-rules/ ; https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act ; https://eba.europa.eu/sites/default/files/2025-11/d8b999ce-a1d9-4964-9606-971bbc2aaf89/AI%20Act%20implications%20for%20the%20EU%20banking%20sector.pdf

United Kingdom - no AI Act, a live testing lane and a model-risk rulebook

The UK has no AI Act. Banks answer to existing duties. The PRA’s supervisory statement SS1/23 sets model risk management principles for banks, and a firm using AI in a model-driven decision should read it as the starting point. The FCA runs AI Live Testing, a supervised route for firms to test AI applications. In April 2026 it named a second cohort of eight firms, including Barclays, Experian, Lloyds Banking Group (Scottish Widows) and UBS. On 8 October 2026 the ICO published a report saying ten foundation model developers have made or committed to data protection changes, and opened a call for evidence on agentic AI. A live testing slot is a supervised test, not an approval to go live.

Sources: https://www.bankofengland.co.uk/prudential-regulation/publication/2023/may/model-risk-management-principles-for-banks-ss ; https://www.fca.org.uk/news/press-releases/fca-announces-second-cohort-ai-live-testing ; https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2026/10/ico-secures-changes-from-leading-ai-developers-as-scrutiny-extends-to-ai-agents/

Germany - BaFin ties AI to DORA

BaFin’s guidance on ICT risks in the use of AI at financial entities is dated 23 January 2026. It is non-mandatory. It helps banks and insurers apply DORA when they use AI, covering ICT risk management and ICT third-party risk, including subcontracting. The practical reading: an AI system is an ICT asset and an ICT dependency. If a vendor model sits in a critical process, the DORA register, exit plan and subcontracting chain apply to it.

Source: https://www.bafin.de/SharedDocs/Downloads/EN/Anlage/dl_Anlage_orientierungshilfe_IKT_Risiken_bei_KI_en.pdf?__blob=publicationFile&v=3

France - the ACPR reference on governance and explainability

The ACPR’s reflection paper on the governance of AI algorithms in the financial sector remains the French reference. It grew out of exploratory work with financial firms that the ACPR launched in March 2019. Its focus is explainability and governance of machine-learning systems. It is a supervisory discussion document, not a binding rule. The ACPR has since published a report on the legal and regulatory impacts of AI in banking, finance and insurance (2025). Expect questions about who validated the model and how its output can be explained to a customer or a supervisor.

Sources: https://acpr.banque-france.fr/fr/publications-et-statistiques/publications/gouvernance-des-algorithmes-dintelligence-artificielle-dans-le-secteur-financier ; https://acpr.banque-france.fr/fr/system/files/2025-07/Rapport_68_F_V3.pdf

Netherlands - the SAFEST principles

De Nederlandsche Bank’s general principles for the use of AI in the financial sector use the acronym SAFEST: soundness, accountability, fairness, ethics, skills and transparency. DNB and the AFM have also published a joint report on the impact of AI on the financial sector and supervision, which says AI use is part of the existing processes and services they already supervise. These are supervisory principles and a supervisory report, not legislation. They give a Dutch bank a clear checklist: who is accountable, what has been tested for soundness and fairness, and whether staff have the skills to challenge the model.

Sources: https://www.dnb.nl/media/jkbip2jc/general-principles-for-the-use-of-artificial-intelligence-in-the-financial-sector.pdf ; https://www.dnb.nl/media/gepbbikm/ai.pdf

Ireland - AI as a supervisory spotlight

The Central Bank of Ireland’s 2026 Regulatory and Supervisory Outlook includes a spotlight on approaching AI from a supervisory perspective and an appendix on the AI landscape. It sets the tone: firms are expected to keep strong governance, risk management and operational resilience, and to act in customers’ best interests. It names bias, misinformation, data loss and “black box” opacity as risks. It is an outlook, not a rulebook, but it tells Irish firms where supervisors will look.

Source: https://www.centralbank.ie/docs/default-source/publications/regulatory-and-supervisory-outlook-reports/regulatory-supervisory-outlook-report-2026.pdf

Switzerland - FINMA guidance on AI governance

Switzerland is outside the EU, so the AI Act does not apply directly. FINMA published Guidance 08/2024 on governance and risk management when using AI on 18 December 2024. It describes what FINMA has seen in supervision. It names model risks such as lack of robustness, correctness, explainability and bias, data risks, IT and cyber risks, third-party dependencies, and legal and reputational risks. FINMA observed that most institutions were still early in building their governance. A Swiss bank selling into the EU still has to meet the AI Act for those customers.

Source: https://www.finma.ch/en/news/2024/12/20241218-mm-finma-am-08-24/

Luxembourg - a documentation set from the CSSF

The CSSF has built a set of supporting documents on AI for regulated entities: a white paper on opportunities, risks and recommendations (2018), questions and answers (1 August 2024) and and a 2024 thematic report on AI, all listed on the CSSF’s AI page. This is guidance and observation, not a separate approval regime. Read the thematic report for what the CSSF observed.

Sources: https://www.cssf.lu/en/artificial-intelligence/

Italy - a national AI law that names the financial supervisors

Italy passed its own AI law, Law 132 of 23 September 2025, in force from 10 October 2025. The text keeps the Bank of Italy, Consob and IVASS as the market surveillance authorities for AI systems used in their sectors, so Italian financial AI sits under supervisors the bank already knows. The Bank of Italy has also published a report with the OECD on AI in Italian financial markets (April 2026). Read the law with the EU AI Act, not instead of it.

Sources: https://www.normattiva.it/eli/id/2025/09/25/25G00143/CONSOLIDATED ; https://www.bancaditalia.it/media/notizie/2026/AI-in-Italian-Financial-Markets-en.pdf?language_id=1

Spain - a dedicated AI supervisory agency

Spain has a standalone AI supervisor, the Spanish Agency for the Supervision of Artificial Intelligence (AESIA). For financial firms, the Governor of the Banco de España has said the AI Act gives the national authority responsible for financial supervision the market surveillance role for high-risk financial models. A Spanish bank should expect to deal with AESIA on general AI Act questions and with the Banco de España on financial models.

Sources: https://aesia.digital.gob.es/ ; https://www.bde.es/f/webbe/GAP/Secciones/SalaPrensa/IntervencionesPublicas/Gobernador/Arc/Fic/IIPP-2024-10-07-escriva-en-tr.pdf

The Nordics - supervisors lead with surveys, recommendations and sandboxes

Sweden: Finansinspektionen reported in December 2024 that AI use is spreading quickly in the financial sector while risk management is lagging behind. Denmark: the Danish FSA published recommendations on supervised machine learning in 2019 and an updated set on AI in 2024. Norway: Finanstilsynet ran a regulatory sandbox project on AI in a digital pension adviser and warned in March 2025 against using AI for investment decisions. Finland: FIN-FSA published a thematic review of AI use in the financial sector on 27 June 2025, and a government proposal on AI Act supervision went to Parliament in May 2025. None of these is a binding AI rule. All show what the supervisor will ask.

Sources: https://www.fi.se/en/published/reports/reports/2024/ai-increasingly-common-in-the-financial-sector-but-risk-management-is-lagging-behind/ ; https://www.dfsa.dk/financial-themes/fintech/guidance-and-handheld-supervision/recommendations-when-using-artificial-intelligence ; https://www.finanstilsynet.no/nyhetsarkiv/nyheter/2025/finanstilsynet-advarer-mot-a-bruke-kunstig-intelligens-til-finansielle-investeringer/ ; https://www.finanssivalvonta.fi/en/publications-and-press-releases/supervision-releases/2025/thematic-review-of-the-use-of-AI-in-the-financi

Austria - existing technology-neutral rules, DORA and the AI Act

The FMA says supervision of AI use is already within its remit. AI used for regulated services must meet existing governance, conduct and distribution rules and DORA, and the management body keeps full responsibility for ICT risk, including risk from AI use and from AI-assisted cyber attacks. The AI Act adds to that. There is no separate Austrian AI approval step.

Source: https://www.fma.gv.at/en/cross-sectoral-topics/artificial-intelligence/

Poland - cyber resilience recommendations on advanced AI

Reported, not regulator-confirmed: In late July 2026 the UKNF, the KNF’s supervisory office, issued recommendations to financial market firms on advanced AI models. They say AI can change the cyber threat landscape and raise pressure on digital operational resilience. This comes from CyberDefence24 coverage and has not been checked against the UKNF’s own page, so treat the detail as secondary until confirmed.

Source: https://cyberdefence24.pl/cyberbezpieczenstwo/uknf-sektor-finansowy-musi-byc-gotowy-na-zaawansowana-ai

Belgium and Portugal - no AI-specific bank instrument found

The sources checked for this issue show no AI-specific bank rule from the National Bank of Belgium, the FSMA, the Banco de Portugal or Portuguese supervisors. The NBB runs a fintech and innovation page and the Banco de Portugal runs a Fintech+ contact channel. That is a scoped finding, not proof that nothing exists. Both banks still carry the AI Act, DORA and existing governance duties.

Sources: https://www.nbb.be/en/financial-supervision/fintech-insurtech-innovation ; https://www.bportugal.pt/page/fintech?mlid=2852

Why this matters beyond Europe

Europe shows what happens when a binding horizontal law meets existing sector rules. The law is not the whole map. A later date for high-risk AI in the EU does not delay DORA, model-risk expectations, consumer-credit duties or supervisory questions on governance. Compare it with the earlier issues. Kenya proposed an approval gate. Qatar has two approval triggers. APAC is mostly consultation and guidance. Europe has a fixed legislative timetable and a long supervisory tail. Each answer to “who approved this for production?” is different.

What bank boards should do now

1. Keep building the AI inventory and the high-risk classification, especially for credit scoring. The 2 December 2027 date is a deadline for the Act’s high-risk duties, not a pause on supervisory scrutiny.

2. Map each AI system to DORA: register entry, critical function, vendor, subcontractors and exit plan.

3. Record who validated each model, who challenged it and who can switch it off. This is the common question from the PRA, the ACPR, DNB and FINMA.

4. Treat a sandbox or live-testing slot as a test, not as approval for production.

5. Watch 2 December 2026 for AI-generated content marking and 2 August 2027 for national sandboxes.

Next issue: Latin America.

Source and status note

Positions checked against regulator and institutional materials available on 10 and 11 October 2026. The Digital Omnibus is cited from EUR-Lex (Official Journal, 24 July 2026, in force) and the Council’s final adoption release. Poland is from press coverage and is not yet confirmed from the UKNF. Belgium and Portugal are scoped findings. The CSSF page was last updated in 2024. Several national documents are older supervisory papers and may have been superseded. This is a market map, not legal advice.

THE AI REGULATION MARKET MAP • EUROPE • RAW COMPLIANCE

Eighteen markets. One EU rulebook that moved, and a long tail of supervisors still asking their own questions.

Oonagh van den Berg | RAW Compliance

Issue 3 mapped the Middle East, where one model met several different gates. Europe looks more uniform from a distance. There is one AI Act. Up close, it is a layered map: the AI Act sets the horizontal rule, DORA and the banking rulebook already apply, and national supervisors add their own expectations about governance, explainability and third parties.

The headline for bank boards is a date that moved. The EU has now postponed the high-risk obligations that matter most to banks. That is a timing change, not a reason to stop building controls.

European Union - the AI Act, with a later date for high-risk

The AI Act entered into force on 1 August 2024. The Digital Omnibus on AI changed part of its timetable. The Council and Parliament reached a provisional agreement on 7 May 2026, Parliament gave final approval on 16 June and the Council adopted the text on 29 June. The amending text is Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026 and in force. It delays the high-risk obligations only. Standalone high-risk systems, which include creditworthiness and credit scoring of natural persons, now apply from 2 December 2027 instead of 2 August 2026. High-risk systems embedded in regulated products apply from 2 August 2028. The Council describes the delay as applying to the high-risk rules, not to the whole Act. The Council also lists national AI sandboxes due by 2 August 2027, a grace period for marking AI-generated content that ends on 2 December 2026, and a new prohibition on AI that generates non-consensual sexual content or child sexual abuse material.

The European Banking Authority’s banking-sector paper, updated in November 2025, makes two points that still stand. Creditworthiness assessment of natural persons is high-risk under the Act. And CRD, CRR, DORA, the consumer and mortgage credit directives, the payment services rules and EBA guidelines already apply to banks using AI. Delay does not remove those duties.

Sources: https://eur-lex.europa.eu/eli/reg/2026/1744/oj/eng ; https://www.consilium.europa.eu/en/press/press-releases/2026/06/29/artificial-intelligence-council-gives-final-green-light-to-simplify-and-streamline-rules/ ; https://www.consilium.europa.eu/en/press/press-releases/2026/05/07/artificial-intelligence-council-and-parliament-agree-to-simplify-and-streamline-rules/ ; https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act ; https://eba.europa.eu/sites/default/files/2025-11/d8b999ce-a1d9-4964-9606-971bbc2aaf89/AI%20Act%20implications%20for%20the%20EU%20banking%20sector.pdf

United Kingdom - no AI Act, a live testing lane and a model-risk rulebook

The UK has no AI Act. Banks answer to existing duties. The PRA’s supervisory statement SS1/23 sets model risk management principles for banks, and a firm using AI in a model-driven decision should read it as the starting point. The FCA runs AI Live Testing, a supervised route for firms to test AI applications. In April 2026 it named a second cohort of eight firms, including Barclays, Experian, Lloyds Banking Group (Scottish Widows) and UBS. On 8 October 2026 the ICO published a report saying ten foundation model developers have made or committed to data protection changes, and opened a call for evidence on agentic AI. A live testing slot is a supervised test, not an approval to go live.

Sources: https://www.bankofengland.co.uk/prudential-regulation/publication/2023/may/model-risk-management-principles-for-banks-ss ; https://www.fca.org.uk/news/press-releases/fca-announces-second-cohort-ai-live-testing ; https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2026/10/ico-secures-changes-from-leading-ai-developers-as-scrutiny-extends-to-ai-agents/

Germany - BaFin ties AI to DORA

BaFin’s guidance on ICT risks in the use of AI at financial entities is dated 23 January 2026. It is non-mandatory. It helps banks and insurers apply DORA when they use AI, covering ICT risk management and ICT third-party risk, including subcontracting. The practical reading: an AI system is an ICT asset and an ICT dependency. If a vendor model sits in a critical process, the DORA register, exit plan and subcontracting chain apply to it.

Source: https://www.bafin.de/SharedDocs/Downloads/EN/Anlage/dl_Anlage_orientierungshilfe_IKT_Risiken_bei_KI_en.pdf?__blob=publicationFile&v=3

France - the ACPR reference on governance and explainability

The ACPR’s reflection paper on the governance of AI algorithms in the financial sector remains the French reference. It grew out of exploratory work with financial firms that the ACPR launched in March 2019. Its focus is explainability and governance of machine-learning systems. It is a supervisory discussion document, not a binding rule. The ACPR has since published a report on the legal and regulatory impacts of AI in banking, finance and insurance (2025). Expect questions about who validated the model and how its output can be explained to a customer or a supervisor.

Sources: https://acpr.banque-france.fr/fr/publications-et-statistiques/publications/gouvernance-des-algorithmes-dintelligence-artificielle-dans-le-secteur-financier ; https://acpr.banque-france.fr/fr/system/files/2025-07/Rapport_68_F_V3.pdf

Netherlands - the SAFEST principles

De Nederlandsche Bank’s general principles for the use of AI in the financial sector use the acronym SAFEST: soundness, accountability, fairness, ethics, skills and transparency. DNB and the AFM have also published a joint report on the impact of AI on the financial sector and supervision, which says AI use is part of the existing processes and services they already supervise. These are supervisory principles and a supervisory report, not legislation. They give a Dutch bank a clear checklist: who is accountable, what has been tested for soundness and fairness, and whether staff have the skills to challenge the model.

Sources: https://www.dnb.nl/media/jkbip2jc/general-principles-for-the-use-of-artificial-intelligence-in-the-financial-sector.pdf ; https://www.dnb.nl/media/gepbbikm/ai.pdf

Ireland - AI as a supervisory spotlight

The Central Bank of Ireland’s 2026 Regulatory and Supervisory Outlook includes a spotlight on approaching AI from a supervisory perspective and an appendix on the AI landscape. It sets the tone: firms are expected to keep strong governance, risk management and operational resilience, and to act in customers’ best interests. It names bias, misinformation, data loss and “black box” opacity as risks. It is an outlook, not a rulebook, but it tells Irish firms where supervisors will look.

Source: https://www.centralbank.ie/docs/default-source/publications/regulatory-and-supervisory-outlook-reports/regulatory-supervisory-outlook-report-2026.pdf

Switzerland - FINMA guidance on AI governance

Switzerland is outside the EU, so the AI Act does not apply directly. FINMA published Guidance 08/2024 on governance and risk management when using AI on 18 December 2024. It describes what FINMA has seen in supervision. It names model risks such as lack of robustness, correctness, explainability and bias, data risks, IT and cyber risks, third-party dependencies, and legal and reputational risks. FINMA observed that most institutions were still early in building their governance. A Swiss bank selling into the EU still has to meet the AI Act for those customers.

Source: https://www.finma.ch/en/news/2024/12/20241218-mm-finma-am-08-24/

Luxembourg - a documentation set from the CSSF

The CSSF has built a set of supporting documents on AI for regulated entities: a white paper on opportunities, risks and recommendations (2018), questions and answers (1 August 2024) and and a 2024 thematic report on AI, all listed on the CSSF’s AI page. This is guidance and observation, not a separate approval regime. Read the thematic report for what the CSSF observed.

Sources: https://www.cssf.lu/en/artificial-intelligence/

Italy - a national AI law that names the financial supervisors

Italy passed its own AI law, Law 132 of 23 September 2025, in force from 10 October 2025. The text keeps the Bank of Italy, Consob and IVASS as the market surveillance authorities for AI systems used in their sectors, so Italian financial AI sits under supervisors the bank already knows. The Bank of Italy has also published a report with the OECD on AI in Italian financial markets (April 2026). Read the law with the EU AI Act, not instead of it.

Sources: https://www.normattiva.it/eli/id/2025/09/25/25G00143/CONSOLIDATED ; https://www.bancaditalia.it/media/notizie/2026/AI-in-Italian-Financial-Markets-en.pdf?language_id=1

Spain - a dedicated AI supervisory agency

Spain has a standalone AI supervisor, the Spanish Agency for the Supervision of Artificial Intelligence (AESIA). For financial firms, the Governor of the Banco de España has said the AI Act gives the national authority responsible for financial supervision the market surveillance role for high-risk financial models. A Spanish bank should expect to deal with AESIA on general AI Act questions and with the Banco de España on financial models.

Sources: https://aesia.digital.gob.es/ ; https://www.bde.es/f/webbe/GAP/Secciones/SalaPrensa/IntervencionesPublicas/Gobernador/Arc/Fic/IIPP-2024-10-07-escriva-en-tr.pdf

The Nordics - supervisors lead with surveys, recommendations and sandboxes

Sweden: Finansinspektionen reported in December 2024 that AI use is spreading quickly in the financial sector while risk management is lagging behind. Denmark: the Danish FSA published recommendations on supervised machine learning in 2019 and an updated set on AI in 2024. Norway: Finanstilsynet ran a regulatory sandbox project on AI in a digital pension adviser and warned in March 2025 against using AI for investment decisions. Finland: FIN-FSA published a thematic review of AI use in the financial sector on 27 June 2025, and a government proposal on AI Act supervision went to Parliament in May 2025. None of these is a binding AI rule. All show what the supervisor will ask.

Sources: https://www.fi.se/en/published/reports/reports/2024/ai-increasingly-common-in-the-financial-sector-but-risk-management-is-lagging-behind/ ; https://www.dfsa.dk/financial-themes/fintech/guidance-and-handheld-supervision/recommendations-when-using-artificial-intelligence ; https://www.finanstilsynet.no/nyhetsarkiv/nyheter/2025/finanstilsynet-advarer-mot-a-bruke-kunstig-intelligens-til-finansielle-investeringer/ ; https://www.finanssivalvonta.fi/en/publications-and-press-releases/supervision-releases/2025/thematic-review-of-the-use-of-AI-in-the-financi

Austria - existing technology-neutral rules, DORA and the AI Act

The FMA says supervision of AI use is already within its remit. AI used for regulated services must meet existing governance, conduct and distribution rules and DORA, and the management body keeps full responsibility for ICT risk, including risk from AI use and from AI-assisted cyber attacks. The AI Act adds to that. There is no separate Austrian AI approval step.

Source: https://www.fma.gv.at/en/cross-sectoral-topics/artificial-intelligence/

Poland - cyber resilience recommendations on advanced AI

Reported, not regulator-confirmed: In late July 2026 the UKNF, the KNF’s supervisory office, issued recommendations to financial market firms on advanced AI models. They say AI can change the cyber threat landscape and raise pressure on digital operational resilience. This comes from CyberDefence24 coverage and has not been checked against the UKNF’s own page, so treat the detail as secondary until confirmed.

Source: https://cyberdefence24.pl/cyberbezpieczenstwo/uknf-sektor-finansowy-musi-byc-gotowy-na-zaawansowana-ai

Belgium and Portugal - no AI-specific bank instrument found

The sources checked for this issue show no AI-specific bank rule from the National Bank of Belgium, the FSMA, the Banco de Portugal or Portuguese supervisors. The NBB runs a fintech and innovation page and the Banco de Portugal runs a Fintech+ contact channel. That is a scoped finding, not proof that nothing exists. Both banks still carry the AI Act, DORA and existing governance duties.

Sources: https://www.nbb.be/en/financial-supervision/fintech-insurtech-innovation ; https://www.bportugal.pt/page/fintech?mlid=2852

Why this matters beyond Europe

Europe shows what happens when a binding horizontal law meets existing sector rules. The law is not the whole map. A later date for high-risk AI in the EU does not delay DORA, model-risk expectations, consumer-credit duties or supervisory questions on governance. Compare it with the earlier issues. Kenya proposed an approval gate. Qatar has two approval triggers. APAC is mostly consultation and guidance. Europe has a fixed legislative timetable and a long supervisory tail. Each answer to “who approved this for production?” is different.

What bank boards should do now

1. Keep building the AI inventory and the high-risk classification, especially for credit scoring. The 2 December 2027 date is a deadline for the Act’s high-risk duties, not a pause on supervisory scrutiny.

2. Map each AI system to DORA: register entry, critical function, vendor, subcontractors and exit plan.

3. Record who validated each model, who challenged it and who can switch it off. This is the common question from the PRA, the ACPR, DNB and FINMA.

4. Treat a sandbox or live-testing slot as a test, not as approval for production.

5. Watch 2 December 2026 for AI-generated content marking and 2 August 2027 for national sandboxes.

Next issue: Latin America.

Source and status note

Positions checked against regulator and institutional materials available on 10 and 11 October 2026. The Digital Omnibus is cited from EUR-Lex (Official Journal, 24 July 2026, in force) and the Council’s final adoption release. Poland is from press coverage and is not yet confirmed from the UKNF. Belgium and Portugal are scoped findings. The CSSF page was last updated in 2024. Several national documents are older supervisory papers and may have been superseded. This is a market map, not legal advice.

THE AI REGULATION MARKET MAP • EUROPE • RAW COMPLIANCE