2026 04 24 ai agent regulation global financial services

Global artificial intelligence agent regulation in financial services: non-functional requirement obligations and low-code citizen-development controls

Research Question

What regulatory obligations do financial-services regulators globally, including the European Union (EU), Australia, New Zealand (NZ), the United States (US), and the United Kingdom (UK), impose on Artificial Intelligence (AI) agents and agentic systems used in regulated processes such as credit, insurance, payments, and advice, what cross-cutting control requirements such as explainability, auditability, robustness, and human oversight do those obligations mandate, and how do those requirements apply when business users create and deploy agents using low-code platforms such as Microsoft Copilot Studio?

Scope

In scope:

European Union AI Act (Regulation (EU) 2024/1689): high-risk AI system classification, conformity assessment obligations, technical documentation, human oversight, data governance, and change management requirements for AI in credit, insurance, and financial-advice processes.
Australian Prudential Regulation Authority (APRA) Prudential Practice Guide (CPG) 234 (Information Security) and related APRA governance and operational-risk material: obligations on regulated entities deploying AI in material business processes.
Reserve Bank of New Zealand (RBNZ) named financial-stability concerns attributable to AI: model error, privacy exposure, cyber-exposure, market distortions, and third-party AI concentration risk.
Financial Markets Authority (FMA) supervisory signals on AI governance for NZ-regulated entities.
NZ existing legislation already imposing obligations on AI-enabled processes: Privacy Act 2020, Fair Trading Act 1986 (FTA), and Companies Act 1993.
US federal financial-regulator guidance, primarily the Federal Reserve and Office of the Comptroller of the Currency (OCC) interagency model-risk framework and Consumer Financial Protection Bureau (CFPB) credit-decision guidance.
UK Financial Conduct Authority (FCA) and Prudential Regulation Authority (PRA) AI guidance and the Digital Regulation Cooperation Forum (DRCF) cross-regulator coordination material.
Canadian Office of the Superintendent of Financial Institutions (OSFI) third-party and technology-risk guidance as a comparator.
Cross-cutting control categories imposed across two or more frameworks: explainability, auditability, robustness, fairness and bias controls, security, privacy by design, and human oversight.
Specific implications for low-code citizen-development scenarios: what obligations attach when a business user, rather than an engineer, builds and deploys an AI agent through a platform such as Microsoft Copilot Studio.

Out of scope:

General data-protection law except where it intersects directly with AI-specific or regulated-process obligations.
Technical feature documentation for Microsoft Copilot Studio beyond governance controls relevant to regulated deployment.
Non-financial-services sectors except where a comparator framework directly informs financial-services obligations.
AI model training and fine-tuning techniques unrelated to compliance obligations.

Constraints:

Prioritise published, citable primary sources: legislative text, regulator guidance, consultation papers, enforcement statements, and official framework documents.
Distinguish binding requirements from non-binding supervisory expectations and principles.
All sources must include URLs.
Focus on obligations already in force or scheduled to take effect within 24 months of the research date.

Context

[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai] The EU AI Act already creates an explicit finance-specific AI rule set for some use cases, because Annex III treats creditworthiness evaluation, credit scoring, and life and health insurance risk assessment and pricing as high-risk AI uses and ties them to controls such as risk management, logging, technical documentation, human oversight, and robustness.

[fact; source: https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf; https://www.fma.govt.nz/assets/Research/Understanding-Artificial-Intelligence-in-Financial-Services.pdf; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/; https://www.legislation.govt.nz/act/public/1986/0121/latest/DLM96439.html; https://www.legislation.govt.nz/act/public/1993/0105/1.0/whole.html] Outside the EU, the regulatory picture is more technology-neutral, but it is not permissive, because prudential, privacy, conduct, and directors' duty frameworks already apply to AI-enabled workflows and New Zealand regulators are publicly treating AI as a live supervisory topic.

[inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-24-business-led-low-code-agent-governance.md] The practical governance problem is that low-code agent platforms compress the time between idea and deployment while legal accountability stays with the regulated institution, so firms need to know which controls are mandatory before business users are allowed to ship agents into regulated workflows.

Approach

Map the EU AI Act provisions applicable to financial services and identify when an institution using a low-code platform is acting as provider, deployer, or both.
Review APRA information-security and operational-risk material to identify what Australian entities must already do for AI in material processes.
Map the NZ legislative floor, including privacy, misleading-conduct, and directors' duty obligations, and separate it from non-binding regulator guidance.
Review RBNZ and FMA publications for supervisory signals and explicit AI risk categories.
Survey US federal guidance for model-risk, explainability, and adverse-action obligations.
Review UK FCA and PRA material and DRCF coordination outputs to identify the UK posture toward AI in finance.
Extract the cross-cutting control categories that recur across jurisdictions and convert them into a minimum compliance posture.
Synthesize the specific implications for low-code citizen development and identify the controls required before a business user can legitimately deploy an agent into a regulated process.

Sources

Starting points, papers, articles, and official documents.

Research Skill Output

(Full output from running the research skill, retained verbatim in the completed item. Sections 0 to 5 are the investigation, section 6 seeds the Findings section below.)

§0 Initialise

[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence] Research question restated: which binding and supervisory obligations already govern Artificial Intelligence (AI) agents used in regulated financial-services processes, which cross-cutting control requirements those obligations imply, and how those obligations apply when business users build or deploy agents through low-code tooling.
[fact; source: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai; https://www.fma.govt.nz/assets/Research/Understanding-Artificial-Intelligence-in-Financial-Services.pdf; https://www.osfi-bsif.gc.ca/en/print/pdf/node/1680] Scope confirmed: the investigation covers the European Union (EU), Australia, New Zealand (NZ), the United States (US), the United Kingdom (UK), and Canada as a comparator; it focuses on regulated finance workflows such as credit, insurance, payments, advice, and customer communications; and it treats low-code governance as a deployment-control problem rather than a vendor feature-comparison problem.
[fact; source: https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-24-business-led-low-code-agent-governance.md; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-ai-governance-assurance-change-control-verification.md; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-02-28-rbnz-ai-supervisory-expectations.md] Prior work cross-reference: prior completed items already established that regulated AI deployment depends on change control, explicit routing, and central low-code governance, so this item focuses on the legal and supervisory floor that those operating models must satisfy.
[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://www.consumerfinance.gov/about-us/newsroom/cfpb-issues-guidance-on-credit-denials-by-lenders-using-artificial-intelligence/] Output format: this item produces a knowledge output, namely a cross-jurisdiction regulatory baseline for deploying AI agents in financial services.

§1 Question Decomposition

Root question: What legal and supervisory obligations already apply when a financial institution uses an AI agent in a regulated process, and what does that mean for low-code deployment?
A. EU legal trigger
- A1. Which financial-services AI uses are explicitly high-risk under the EU AI Act?
- A2. Which Articles create the operational controls for those systems?
- A3. When does a deployer or low-code customiser become a provider?
B. Australia prudential baseline
- B1. What does Australian Prudential Regulation Authority (APRA) require for information security, operational resilience, and third-party risk in AI-enabled processes?
- B2. Is there any AI-specific APRA finance rule, or only technology-neutral prudential controls?
C. New Zealand legal floor
- C1. Which Privacy Act obligations apply to AI use today?
- C2. Which Fair Trading Act 1986 (FTA) obligations apply to AI-generated communications and services?
- C3. Which Companies Act duties keep directors accountable for AI-driven decisions?
- C4. What do Reserve Bank of New Zealand (RBNZ) and Financial Markets Authority (FMA) say publicly about AI risk?
D. United States and United Kingdom comparator controls
- D1. What do Supervisory Letter SR 11-7 and Consumer Financial Protection Bureau (CFPB) guidance require for explainability, validation, and governance?
- D2. What is the UK supervisory posture under Discussion Paper 5/22 (DP5/22), Feedback Statement 23/6 (FS23/6), and Digital Regulation Cooperation Forum (DRCF) coordination?
E. Cross-jurisdiction control patterns
- E1. Which cross-cutting control categories recur across at least two frameworks?
- E2. Which controls are binding, which are supervisory expectations, and which are best-practice comparators?
F. Low-code deployment implication
- F1. Does low-code move liability from the institution to the business maker?
- F2. Which approval gates and platform controls are necessary before deployment into a regulated process?

§2 Investigation

A. European Union, the only explicit finance-specific AI rulebook in scope

[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai] Annex III of the EU AI Act treats AI systems used to evaluate the creditworthiness of natural persons or establish their credit score, and AI systems used for risk assessment and pricing in life and health insurance, as high-risk uses.
[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai] High-risk AI systems are subject to obligations covering risk management, data and data-governance quality, technical documentation, record-keeping and logging, transparency to deployers, human oversight, accuracy, robustness, cybersecurity, post-market monitoring, and conformity assessment before placing on the market or putting into service.
[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689] The Act defines a deployer as a person using an AI system under its authority, and it also states that a distributor, importer, deployer, or other third party can become the provider of a high-risk AI system if it puts the system into service under its own name or trademark or makes a substantial modification affecting compliance or intended purpose.
[inference; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance] A financial institution that configures a Copilot Studio-built agent for creditworthiness or insurance-risk decisions could be at least a deployer and, depending on branding and modification, also a provider, so low-code assembly does not remove EU operator obligations.

B. Australia, technology-neutral prudential obligations with direct AI relevance

[fact; source: https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf] APRA's Prudential Practice Guide (CPG) 234 says information assets should be classified by criticality and sensitivity, including assets managed by third parties and related parties, and it recommends lifecycle management so information-security requirements are considered from planning and acquisition through decommissioning and destruction.
[fact; source: https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf] CPG 234 also expects clear accountability, escalation, reporting, and incident-response arrangements, including reporting to boards or other governing bodies and enough procedural detail to reduce improvisation during incidents.
[fact; source: https://www.apra.gov.au/sites/default/files/2025-10/APRA%20Annual%20Report%202024-25.pdf; https://www.apra.gov.au/news-and-publications/apra-chair-john-lonsdale-speech-to-australian-banking-association-0] APRA's current public posture emphasises governance, cyber risk, operational risk, material service-provider risk, and resilience, including heightened requirements under Prudential Standard (CPS) 230 and continuing insistence that institutions meet full CPS 234 obligations.
[inference; source: https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.apra.gov.au/sites/default/files/2025-10/APRA%20Annual%20Report%202024-25.pdf; https://www.apra.gov.au/news-and-publications/apra-chair-john-lonsdale-speech-to-australian-banking-association-0] APRA has not yet published a standalone prudential rule for AI agents in finance, but any AI used in a material process already falls inside APRA's information-security, operational-risk, governance, and third-party control expectations.

C. New Zealand, no single AI finance statute but a real legislative and supervisory floor

[fact; source: https://legislation.govt.nz/act/public/2020/0031/169.0/LMS23376.html; https://www.privacy.org.nz/assets/New-order/Resources-/Publications/Guidance-resources/AI-Guidance-Resources-/AI-and-the-Information-Privacy-Principles.pdf] Privacy Act 2020 information privacy principle 8 requires agencies to take reasonable steps to ensure personal information is accurate, up to date, complete, relevant, and not misleading before use or disclosure, and the Privacy Commissioner applies that principle directly to AI tools.
[fact; source: https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/; https://www.privacy.org.nz/assets/New-order/Resources-/Publications/Guidance-resources/AI-Guidance-Resources-/AI-and-the-Information-Privacy-Principles.pdf] The Office of the Privacy Commissioner says the Privacy Act is technology-neutral and expects agencies using AI tools to obtain senior leadership approval, conduct a Privacy Impact Assessment, be transparent, develop procedures for accuracy and access, ensure human review before acting, and prevent retention or disclosure of personal or confidential information by the tool.
[fact; source: https://legislation.govt.nz/act/public/1986/121/en/2024-11-16.pdf] The FTA prohibits misleading or deceptive conduct generally and misleading conduct in relation to services, so AI-generated financial communications, recommendations, or customer-service outputs that mislead customers can already create statutory exposure.
[fact; source: https://www.legislation.govt.nz/act/public/1993/0105/1.0/whole.html] The Companies Act imposes directors' duties to act in good faith and in the best interests of the company and to exercise the care, diligence, and skill that a reasonable director would exercise, which preserves board accountability for material AI-driven decision systems.
[fact; source: https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf] RBNZ's May 2025 special topic says AI adoption is accelerating, can improve model accuracy and cyber resilience, may amplify AI-driven errors, privacy concerns, market distortions, and cyber attacks, and creates concentration risk through reliance on critical third-party providers.
[fact; source: https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf] The same RBNZ publication states that regulated entities should be aware of and manage potential AI risks in line with their regulatory obligations, which is a supervisory signal rather than a new rule.
[fact; source: https://www.fma.govt.nz/assets/Research/Understanding-Artificial-Intelligence-in-Financial-Services.pdf; https://www.fma.govt.nz/assets/Corporate-Publications/FMA-Annual-Report-2025.pdf] FMA describes itself as a technology-neutral and pro-innovation regulator, treats AI as a changing risk landscape for financial-services firms, and has incorporated AI research into its conduct-regulation agenda and outcomes-focused supervision.
[inference; source: https://legislation.govt.nz/act/public/2020/0031/169.0/LMS23376.html; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/; https://legislation.govt.nz/act/public/1986/121/en/2024-11-16.pdf; https://www.legislation.govt.nz/act/public/1993/0105/1.0/whole.html; https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf] NZ does not yet have a single finance-specific AI statute, but institutions already face enforceable privacy, conduct, and governance duties plus clear supervisory scrutiny.

D. United States, explainability and model-risk controls through existing laws

[fact; source: https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf] SR 11-7, issued by the Federal Reserve with the OCC, requires disciplined model development, strong documentation, independent validation, ongoing monitoring, outcomes analysis, board and senior-management governance, internal audit involvement, and maintenance of a model inventory.
[fact; source: https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf] SR 11-7 also treats model risk as arising from fundamental model error or misuse of model outputs, and it says model-risk management must become more rigorous when model outputs materially influence business decisions or risk management.
[fact; source: https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf; https://www.consumerfinance.gov/about-us/newsroom/cfpb-issues-guidance-on-credit-denials-by-lenders-using-artificial-intelligence/] CFPB states that Equal Credit Opportunity Act and Regulation B adverse-action requirements apply equally to complex algorithms, artificial intelligence, and black-box models, and that creditors cannot use opaque technology as a defence if they cannot provide specific and accurate reasons for adverse actions.
[fact; source: https://www.occ.gov/news-issuances/news-releases/2021/nr-ia-2021-100a.pdf] In 2021, the US federal banking agencies and CFPB jointly requested information on financial institutions' use of AI, which shows coordinated supervisory attention but not a standalone AI-specific finance rulebook.
[inference; source: https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf] In practice, the strongest current US controls on AI agents in regulated finance come from model validation, documentation, monitoring, and adverse-action explainability rather than from a dedicated AI-agent statute.

E. United Kingdom, principles-based supervision and cross-regulator coordination

[fact; source: https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence] DP5/22 says AI in UK financial services can bring benefits but also novel or amplified risks to consumers, firms, market integrity, and financial stability, and it frames the central regulatory question as whether existing rules are sufficient or need clarification.
[fact; source: https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence] DP5/22 further says the supervisory authorities' approach is largely limited to clarifying how the existing regulatory framework applies to AI and addressing identified gaps, with emphasis on governance and decision-making challenges.
[fact; source: https://www.fca.org.uk/publications/feedback-statements/fs23-6-artifical-intelligence-machine-learning] FS23/6 confirms that FCA and the Bank are focused on safe and responsible adoption of AI in financial services and are using the discussion paper to decide whether additional clarification of existing regulations would be useful.
[fact; source: https://www.drcf.org.uk/siteassets/drcf/pdf-files/drcf-annual-report-2023_24?v=383901] DRCF's 2023/24 annual report says UK government AI regulation rests on five cross-sector principles interpreted by regulators within their remits, and that DRCF coordinated work on AI governance, algorithmic systems, generative AI, and the AI and Digital Hub.
[inference; source: https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.fca.org.uk/publications/feedback-statements/fs23-6-artifical-intelligence-machine-learning; https://www.drcf.org.uk/siteassets/drcf/pdf-files/drcf-annual-report-2023_24?v=383901] The UK posture is explicitly principles-based and coordination-heavy, so firms should expect governance, accountability, and documentation obligations to come through existing conduct, prudential, and consumer-protection frameworks rather than through an AI-specific financial-services rule.

F. Canada as a useful third-party and technology-risk comparator

[fact; source: https://www.osfi-bsif.gc.ca/en/print/pdf/node/1680] The Office of the Superintendent of Financial Institutions (OSFI) final B-10 guidance response says B-10 is outcomes-focused, principles-based, and risk-based, applies to an expanded third-party ecosystem, and requires proportional management of subcontractor risk and concentration risk.
[fact; source: https://www.osfi-bsif.gc.ca/en/guidance/guidance-library/osfi-releases-final-guideline-b-13-technology-cyber-risk-management-letter-2022] OSFI's B-13 release letter says B-13 sets expectations for the sound management of technology and cyber risk, should be implemented from a risk-based perspective, and complements B-10, corporate governance, and operational-risk guidance.
[inference; source: https://www.osfi-bsif.gc.ca/en/print/pdf/node/1680; https://www.osfi-bsif.gc.ca/en/guidance/guidance-library/osfi-releases-final-guideline-b-13-technology-cyber-risk-management-letter-2022] Even without AI-specific finance rules, OSFI's framework reinforces the same control class that AI-agent deployments need: accountable governance, third-party due diligence, concentration-risk management, and resilient technology operations.

G. Cross-jurisdiction control categories

[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.osfi-bsif.gc.ca/en/print/pdf/node/1680] Governance and accountability recur across every jurisdiction reviewed, whether through explicit operator duties, board oversight, or regulated-entity accountability for outsourced services and model use.
[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/] Explainability, traceability, and human review recur through the EU AI Act's logging and human-oversight duties, SR 11-7's documentation and validation requirements, CFPB's specific-reasons obligation, and NZ privacy guidance on human review and transparency.
[fact; source: https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.osfi-bsif.gc.ca/en/guidance/guidance-library/osfi-releases-final-guideline-b-13-technology-cyber-risk-management-letter-2022; https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689] Security, resilience, and lifecycle control recur through APRA and OSFI technology-risk guidance and through the EU AI Act's robustness and cybersecurity requirements.
[fact; source: https://legislation.govt.nz/act/public/2020/0031/169.0/LMS23376.html; https://www.privacy.org.nz/assets/New-order/Resources-/Publications/Guidance-resources/AI-Guidance-Resources-/AI-and-the-Information-Privacy-Principles.pdf; https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689] Data quality and privacy governance recur through EU data-governance rules and NZ accuracy and privacy obligations.
[inference; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.osfi-bsif.gc.ca/en/print/pdf/node/1680; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/] The minimum credible compliance posture for AI agents in regulated processes is therefore not a single policy, but a control stack covering use-case classification, named ownership, approved data sources, validation, logging, human review, incident handling, and vendor oversight.

H. Low-code citizen-development implications

[fact; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance] Microsoft Copilot Studio provides governance controls including data policies over knowledge sources, actions, triggers, Hypertext Transfer Protocol (HTTP) requests, and publication; maker audit logs in Microsoft Purview; audit and alerting support in Microsoft Sentinel; sensitivity labels on some sources; security warnings before publishing; environment routing; and the ability for administrators to disable publishing.
[fact; source: https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/; https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689] Official frameworks place responsibility on the regulated organisation, not the citizen developer, because NZ privacy guidance expects senior-leadership approval and human review, while the EU AI Act allocates duties to providers and deployers rather than to individual makers as such.
[inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf] Low-code platform controls are useful compliance enablers, but they do not replace legal controls such as risk classification, model validation, human oversight, documentation, or adverse-action explainability.
[inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-24-business-led-low-code-agent-governance.md] A business user can safely self-serve only for low-impact internal support uses, whereas deployment into credit, insurance, regulated advice, or customer-facing decisions requires a central approval gate and institution-level accountability.

§3 Reasoning

[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689] The EU answer is straightforward because the AI Act names the relevant finance use cases and prescribes the control categories directly.
[inference; source: https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.osfi-bsif.gc.ca/en/print/pdf/node/1680] The non-EU answer must be derived from technology-neutral frameworks, but the control implications are still concrete because those frameworks already govern information security, model risk, operational resilience, third-party dependence, and management accountability.
[fact; source: https://legislation.govt.nz/act/public/2020/0031/169.0/LMS23376.html; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/] The NZ privacy layer is narrower than an EU-style automated-decision right, because the binding statute emphasises information handling and accuracy before use or disclosure, while the Privacy Commissioner fills in AI-specific expectations through guidance rather than through a new statutory right against automated decisions.
[inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf] The low-code question is therefore not whether business users may build agents, but whether the institution has inserted enough governance between build and deployment to satisfy the same duties that would apply if engineers had built the system.

§4 Consistency Check

[fact; source: https://legislation.govt.nz/act/public/2020/0031/169.0/LMS23376.html; https://www.privacy.org.nz/assets/New-order/Resources-/Publications/Corporate-reports/2024-annual-report/2024-26-11-We-need-Privacy-Act-modernisation-Annual-Report-2024.pdf] The main correction surfaced during consistency review is that NZ law does not currently provide a strong standalone automated-decision right comparable to some overseas privacy regimes, because the Privacy Commissioner is publicly asking for stronger automated-decision protections as a future reform.
[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai] No contradiction remained in the EU section after checking the Commission overview against the AI Act text, because both sources identify credit scoring and life and health insurance risk assessment as high-risk and list the same core control categories.
[fact; source: https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.fca.org.uk/publications/feedback-statements/fs23-6-artifical-intelligence-machine-learning] No contradiction remained in the UK section after checking DP5/22 and FS23/6, because both point toward clarification of existing frameworks rather than immediate sector-specific AI rulemaking.

§5 Depth and Breadth Expansion

[inference; source: https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf; https://www.osfi-bsif.gc.ca/en/print/pdf/node/1680; https://www.apra.gov.au/news-and-publications/apra-chair-john-lonsdale-speech-to-australian-banking-association-0] Systemic-risk lens: concentration risk matters as much as model quality, because regulators repeatedly emphasise critical third-party dependence and operational resilience when digital infrastructure becomes shared and concentrated.
[inference; source: https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/] Operational lens: the real work of compliance happens after model selection, in validation, documentation, human review, exception handling, and adverse-decision explanation.
[inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-24-business-led-low-code-agent-governance.md] Behavioural lens: low-code broadens the maker population and therefore increases the probability of bypassing controls, which means platform convenience raises the value of publishing gates, environment controls, and maker monitoring rather than reducing it.
[inference; source: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai; https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.fca.org.uk/publications/feedback-statements/fs23-6-artifical-intelligence-machine-learning] Regulatory-evolution lens: jurisdictions tend to start with technology-neutral supervision and then add more explicit AI obligations only where rights impacts or market-harm risks justify it, which is why the EU is prescriptive first and the UK, Australia, and NZ remain principles-led.

§6 Synthesis

Executive summary:

[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/; https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf; https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence/] Financial institutions already face enforceable obligations when they deploy AI agents in regulated workflows, because the EU AI Act imposes explicit high-risk controls for some finance use cases while the US, Australia, NZ, and the UK already apply model-risk, privacy, conduct, operational-risk, and governance rules to the same underlying activities. [inference; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai; https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence] Compared with the jurisdictions reviewed outside the EU, the EU is the most prescriptive regime in scope, because creditworthiness and life and health insurance risk-assessment uses are high-risk and require risk management, logging, documentation, human oversight, robustness, and conformity assessment. [inference; source: https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf; https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.osfi-bsif.gc.ca/en/print/pdf/node/1680] Outside the EU, the rule pattern is technology-neutral but still demanding, because regulators expect secure information handling, accountable governance, validation, resilience, vendor oversight, and human review even where they have not issued an AI-specific finance code. [inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/] Low-code citizen development does not shift responsibility away from the institution, so business-built agents in regulated processes must still pass central approval, documentation, logging, testing, and oversight gates before deployment.

Key findings:

High confidence. [fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai] The EU AI Act already makes AI systems used for creditworthiness evaluation, credit scoring, and life and health insurance risk assessment and pricing high-risk, which means those systems cannot lawfully be deployed without documented risk management, logging, technical documentation, human oversight, robustness, cybersecurity, and conformity-assessment controls.
High confidence. [inference; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance] A regulated institution using a low-code platform to configure an agent for a high-risk financial use case remains at least a deployer under the EU AI Act and may also become a provider through own-branding or substantial modification, so low-code assembly does not reduce operator obligations.
Medium confidence. [inference; source: https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.apra.gov.au/sites/default/files/2025-10/APRA%20Annual%20Report%202024-25.pdf] APRA has not published a standalone AI prudential standard for financial services, but its existing information-security and operational-risk framework would require AI used in material processes to sit inside classified information-asset inventories, lifecycle security controls, board reporting, incident response, and material service-provider oversight.
High confidence. [fact; source: https://legislation.govt.nz/act/public/2020/0031/169.0/LMS23376.html; https://legislation.govt.nz/act/public/1986/121/en/2024-11-16.pdf; https://www.legislation.govt.nz/act/public/1993/0105/1.0/whole.html; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/] NZ already imposes a real legal floor on AI deployments through privacy, misleading-conduct, and directors' duties law, and the Office of the Privacy Commissioner adds official expectations for leadership approval, Privacy Impact Assessment, transparency, human review, and controls over retention and disclosure.
High confidence. [fact; source: https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf; https://www.fma.govt.nz/assets/Research/Understanding-Artificial-Intelligence-in-Financial-Services.pdf] RBNZ and FMA have moved AI into active supervisory attention by naming AI-driven errors, privacy and cyber harms, market distortions, concentration risk, and conduct challenges as current concerns, even though they have not yet converted those concerns into a dedicated finance-specific AI rulebook.
High confidence. [fact; source: https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf] The strongest current US obligations for AI in regulated financial decisions come from model-risk governance and adverse-action explainability, because SR 11-7 requires documented validation and board governance while CFPB says creditors may not use opaque models if they cannot provide specific and accurate reasons for denials.
Medium confidence. [inference; source: https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.fca.org.uk/publications/feedback-statements/fs23-6-artifical-intelligence-machine-learning; https://www.drcf.org.uk/siteassets/drcf/pdf-files/drcf-annual-report-2023_24?v=383901] The UK is unlikely in the near term to create a separate financial-services AI code equivalent to the EU AI Act, because the supervisory direction remains to clarify and coordinate existing principles-based regimes rather than replace them with a new sector-specific AI statute.
Medium confidence. [inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-24-business-led-low-code-agent-governance.md] A defensible minimum control set for business-built AI agents in regulated workflows includes a central approval gate with risk classification, named accountability, approved data sources, validation and testing, logging, human review, incident handling, vendor due diligence, and restricted publishing, because platform guardrails alone do not satisfy the underlying legal duties.

Evidence map:

Claim	Source	Confidence	Notes
[fact] EU creditworthiness and life and health insurance AI uses are high-risk and must meet Title III controls.	https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai	high	Binding legal rule.
[inference] Low-code institutions remain deployers and can become providers through own-branding or substantial modification.	https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance	high	Legal role analysis applied to low-code deployment.
[inference] APRA's current framework would capture AI through information-security and operational-risk controls.	https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.apra.gov.au/sites/default/files/2025-10/APRA%20Annual%20Report%202024-25.pdf	medium	Technology-neutral prudential baseline interpreted for AI use in material processes.
[fact] NZ already has a statutory AI floor through privacy, misleading-conduct, and directors' duties law, plus privacy guidance.	https://legislation.govt.nz/act/public/2020/0031/169.0/LMS23376.html; https://legislation.govt.nz/act/public/1986/121/en/2024-11-16.pdf; https://www.legislation.govt.nz/act/public/1993/0105/1.0/whole.html; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/	high	Binding statutes plus official but non-binding Privacy Commissioner expectations.
[fact] RBNZ and FMA are already treating AI as a supervisory topic through risk publications and conduct research.	https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf; https://www.fma.govt.nz/assets/Research/Understanding-Artificial-Intelligence-in-Financial-Services.pdf	high	Supervisory signal, not new rulemaking.
[fact] US AI obligations in finance currently centre on model validation and adverse-action explainability.	https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf	high	Interagency model-risk guidance plus CFPB policy statement.
[inference] UK policy remains principles-based and coordination-led rather than prescriptive like the EU.	https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.fca.org.uk/publications/feedback-statements/fs23-6-artifical-intelligence-machine-learning; https://www.drcf.org.uk/siteassets/drcf/pdf-files/drcf-annual-report-2023_24?v=383901	medium	Strong official signal, but future legislation can still change.
[inference] Business-built agents in regulated workflows need central approval, validation, logging, and publishing controls before deployment.	https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf	high	Derived minimum control posture across frameworks.

Assumptions:

[assumption] The low-code scenarios considered here involve agents that influence or participate in regulated financial workflows rather than purely personal productivity tasks. Justification: the research question is limited to credit, insurance, payments, advice, and related regulated processes, so the control analysis assumes consequential use rather than casual drafting.

Analysis:

[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai] The legal evidence is strongest in the EU because the Act identifies the relevant finance use cases and names the operator obligations directly. [inference; source: https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.osfi-bsif.gc.ca/en/print/pdf/node/1680] Outside the EU, the correct analytical move is to map AI agents onto existing categories such as models, information assets, critical operations, and third-party arrangements rather than to search for the phrase "AI agent" in current prudential rules. [fact; source: https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf; https://legislation.govt.nz/act/public/1986/121/en/2024-11-16.pdf] Conduct and consumer-protection risk also remain central because explainability and non-misleading communication duties bite at the point where an AI output affects a customer, not only at the point where a model is built. [inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/] That is why low-code governance must be judged by whether it creates evidence, accountability, and control over deployment, not by whether the platform advertises secure defaults.

Risks, gaps, uncertainties:

[inference; source: https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf] The RBNZ portion relies heavily on a single primary publication, so a future review should corroborate it with additional RBNZ material if more AI-specific speeches or supervisory statements are published.
[inference; source: https://www.apra.gov.au/sites/default/files/2025-10/APRA%20Annual%20Report%202024-25.pdf; https://www.apra.gov.au/news-and-publications/apra-chair-john-lonsdale-speech-to-australian-banking-association-0] APRA may still develop more explicit AI-governance expectations, but current public material does not yet amount to a dedicated AI prudential standard.
[inference; source: https://www.fca.org.uk/publications/feedback-statements/fs23-6-artifical-intelligence-machine-learning; https://www.drcf.org.uk/siteassets/drcf/pdf-files/drcf-annual-report-2023_24?v=383901] UK coordination work is active and could harden into clearer cross-regulator expectations, so the current principles-based reading should be treated as time-sensitive rather than permanent.
[inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance] Microsoft platform controls were assessed only from public governance documentation, not through a tenant-level product test, so the analysis covers control availability and governance logic rather than implementation quality in a particular environment.

Open questions:

[inference; source: https://www.fma.govt.nz/assets/Corporate-Publications/FMA-Annual-Report-2025.pdf] How should NZ's Conduct of Financial Institutions regime be mapped explicitly onto AI-assisted financial-advice and sales workflows?
[inference; source: https://www.occ.gov/news-issuances/news-releases/2021/nr-ia-2021-100a.pdf] Which US agencies are most likely to move next from general AI inquiry into finance-specific supervisory expectations for agentic workflows?
[inference; source: https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.drcf.org.uk/siteassets/drcf/pdf-files/drcf-annual-report-2023_24?v=383901] Will the UK eventually turn DRCF coordination and DP5/22 themes into a more explicit assurance regime for high-impact financial AI systems?

§7 Recursive Review

Status: completed.
Checks applied: acronym first-use audit, inline claim-label audit, Context audit, Evidence Map audit, banned-phrase audit, and em-dash audit.
[fact; source: https://legislation.govt.nz/act/public/2020/0031/latest/LMS23223.html; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/] Main factual correction: the review removed any implication that NZ already has a standalone automated-decision right equivalent to more prescriptive foreign privacy regimes.
Residual uncertainties are confined to the Risks, Gaps, and Uncertainties section.

Findings

Executive Summary

[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/; https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf; https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence/] Financial institutions already face enforceable obligations when they deploy AI agents in regulated workflows, because the EU AI Act imposes explicit high-risk controls for some finance use cases while the US, Australia, NZ, and the UK already apply model-risk, privacy, conduct, operational-risk, and governance rules to the same underlying activities. [inference; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai; https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence] Compared with the jurisdictions reviewed outside the EU, the EU is the most prescriptive regime in scope, because creditworthiness and life and health insurance risk-assessment uses are treated as high-risk and must meet risk management, logging, documentation, human-oversight, robustness, and conformity-assessment duties. [inference; source: https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf; https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.osfi-bsif.gc.ca/en/print/pdf/node/1680] Outside the EU, regulators mostly rely on technology-neutral frameworks, but those frameworks still require secure information handling, accountable governance, validation, resilience, vendor oversight, and human review for consequential AI uses. [inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/] Low-code citizen development does not shift accountability away from the institution, so business-built agents in regulated processes must still pass central approval, testing, logging, documentation, and oversight gates before deployment.

Key Findings

High confidence. [fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai] The EU AI Act already makes AI systems used for creditworthiness evaluation, credit scoring, and life and health insurance risk assessment and pricing high-risk, which means those systems cannot lawfully be deployed without documented risk management, logging, technical documentation, human oversight, robustness, cybersecurity, and conformity-assessment controls.
High confidence. [inference; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance] A regulated institution using a low-code platform to configure an agent for a high-risk financial use case remains at least a deployer under the EU AI Act and may also become a provider through own-branding or substantial modification, so low-code assembly does not reduce operator obligations.
Medium confidence. [inference; source: https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.apra.gov.au/sites/default/files/2025-10/APRA%20Annual%20Report%202024-25.pdf] APRA has not published a standalone AI prudential standard for financial services, but its existing information-security and operational-risk framework would require AI used in material processes to sit inside classified information-asset inventories, lifecycle security controls, board reporting, incident response, and material service-provider oversight.
High confidence. [fact; source: https://legislation.govt.nz/act/public/2020/0031/169.0/LMS23376.html; https://legislation.govt.nz/act/public/1986/121/en/2024-11-16.pdf; https://www.legislation.govt.nz/act/public/1993/0105/1.0/whole.html; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/] NZ already imposes a real legal floor on AI deployments through privacy, misleading-conduct, and directors' duties law, and the Office of the Privacy Commissioner adds official expectations for leadership approval, Privacy Impact Assessment, transparency, human review, and controls over retention and disclosure.
High confidence. [fact; source: https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf; https://www.fma.govt.nz/assets/Research/Understanding-Artificial-Intelligence-in-Financial-Services.pdf] RBNZ and FMA have moved AI into active supervisory attention by naming AI-driven errors, privacy and cyber harms, market distortions, concentration risk, and conduct challenges as current concerns, even though they have not yet converted those concerns into a dedicated finance-specific AI rulebook.
High confidence. [fact; source: https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf] The strongest current US obligations for AI in regulated financial decisions come from model-risk governance and adverse-action explainability, because SR 11-7 requires documented validation and board governance while CFPB says creditors may not use opaque models if they cannot provide specific and accurate reasons for denials.
Medium confidence. [inference; source: https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.fca.org.uk/publications/feedback-statements/fs23-6-artifical-intelligence-machine-learning; https://www.drcf.org.uk/siteassets/drcf/pdf-files/drcf-annual-report-2023_24?v=383901] The UK is unlikely in the near term to create a separate financial-services AI code equivalent to the EU AI Act, because the supervisory direction remains to clarify and coordinate existing principles-based regimes rather than replace them with a new sector-specific AI statute.
Medium confidence. [inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-24-business-led-low-code-agent-governance.md] A defensible minimum control set for business-built AI agents in regulated workflows includes a central approval gate with risk classification, named accountability, approved data sources, validation and testing, logging, human review, incident handling, vendor due diligence, and restricted publishing, because platform guardrails alone do not satisfy the underlying legal duties.

Evidence Map

Claim	Source	Confidence	Notes
[fact] EU creditworthiness and life and health insurance AI uses are high-risk and must meet Title III controls.	https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai	high	Binding legal rule.
[inference] Low-code institutions remain deployers and can become providers through own-branding or substantial modification.	https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance	high	Legal role analysis applied to low-code deployment.
[inference] APRA's current framework would capture AI through information-security and operational-risk controls.	https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.apra.gov.au/sites/default/files/2025-10/APRA%20Annual%20Report%202024-25.pdf	medium	Technology-neutral prudential baseline interpreted for AI use in material processes.
[fact] NZ already has a statutory AI floor through privacy, misleading-conduct, and directors' duties law, plus privacy guidance.	https://legislation.govt.nz/act/public/2020/0031/169.0/LMS23376.html; https://legislation.govt.nz/act/public/1986/121/en/2024-11-16.pdf; https://www.legislation.govt.nz/act/public/1993/0105/1.0/whole.html; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/	high	Binding statutes plus official but non-binding Privacy Commissioner expectations.
[fact] RBNZ and FMA are already treating AI as a supervisory topic through risk publications and conduct research.	https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf; https://www.fma.govt.nz/assets/Research/Understanding-Artificial-Intelligence-in-Financial-Services.pdf	high	Supervisory signal, not new rulemaking.
[fact] US AI obligations in finance currently centre on model validation and adverse-action explainability.	https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf	high	Interagency model-risk guidance plus CFPB policy statement.
[inference] UK policy remains principles-based and coordination-led rather than prescriptive like the EU.	https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.fca.org.uk/publications/feedback-statements/fs23-6-artifical-intelligence-machine-learning; https://www.drcf.org.uk/siteassets/drcf/pdf-files/drcf-annual-report-2023_24?v=383901	medium	Strong official signal, but future legislation can still change.
[inference] Business-built agents in regulated workflows need central approval, validation, logging, and publishing controls before deployment.	https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/; https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf	medium	Derived minimum control posture across frameworks.

Assumptions

Assumption: [assumption] The low-code scenarios considered here involve agents that influence or participate in regulated financial workflows rather than purely personal productivity tasks. Justification: [assumption] The research question is limited to credit, insurance, payments, advice, and related regulated processes, so the control analysis assumes consequential use rather than casual drafting.

Analysis

[fact; source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai] The EU position required the least inference because the Act names the relevant finance use cases and the mandatory control categories directly. [inference; source: https://www.federalreserve.gov/supervisionreg/srletters/sr1107.pdf; https://www.apra.gov.au/sites/default/files/cpg_234_information_security_june_2019_1.pdf; https://www.osfi-bsif.gc.ca/en/print/pdf/node/1680] The other jurisdictions were evaluated by mapping AI agents onto pre-existing regulatory objects such as models, information assets, critical operations, and third-party arrangements, which is the correct analytical move where regulators remain technology-neutral. [fact; source: https://files.consumerfinance.gov/f/documents/cfpb_2022-03_circular_2022-05.pdf; https://legislation.govt.nz/act/public/1986/121/en/2024-11-16.pdf] The conduct layer matters as much as the prudential layer because opaque or misleading outputs can breach law at the point they affect a consumer, even when the model build process itself appears controlled. [inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/generative-artificial-intelligence/] That weighting leads to the practical conclusion that low-code governance succeeds only if the institution can prove who approved the use case, what data was allowed, how outputs were checked, and why the agent was safe to publish.

Risks, Gaps, and Uncertainties

[inference; source: https://www.rbnz.govt.nz/-/media/project/sites/rbnz/files/publications/financial-stability-reports/2025/may/special-topic_rise-of-the-machine.pdf] The RBNZ portion relies heavily on a single primary publication, so a future review should corroborate it with additional RBNZ material if more AI-specific speeches or supervisory statements are published.
[inference; source: https://www.apra.gov.au/sites/default/files/2025-10/APRA%20Annual%20Report%202024-25.pdf; https://www.apra.gov.au/news-and-publications/apra-chair-john-lonsdale-speech-to-australian-banking-association-0] APRA may still publish more explicit AI material, but current public evidence does not yet amount to a dedicated AI prudential standard.
[inference; source: https://www.fca.org.uk/publications/feedback-statements/fs23-6-artifical-intelligence-machine-learning; https://www.drcf.org.uk/siteassets/drcf/pdf-files/drcf-annual-report-2023_24?v=383901] UK coordination work is active and could harden into clearer assurance expectations, so the current principles-based reading should be treated as time-sensitive.
[inference; source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance] Microsoft platform controls were assessed from public governance documentation rather than a tenant-level implementation test, so this item covers control availability and governance logic rather than implementation quality in a specific environment.

Open Questions

[inference; source: https://www.fma.govt.nz/assets/Corporate-Publications/FMA-Annual-Report-2025.pdf] How should NZ's Conduct of Financial Institutions regime be mapped explicitly onto AI-assisted financial-advice and sales workflows?
[inference; source: https://www.occ.gov/news-issuances/news-releases/2021/nr-ia-2021-100a.pdf] Which US agencies are most likely to move next from general AI inquiry into finance-specific supervisory expectations for agentic workflows?
[inference; source: https://www.bankofengland.co.uk/prudential-regulation/publication/2022/october/artificial-intelligence; https://www.drcf.org.uk/siteassets/drcf/pdf-files/drcf-annual-report-2023_24?v=383901] Will the UK eventually turn DRCF coordination and DP5/22 themes into a more explicit assurance regime for high-impact financial AI systems?