The six major AP regulators — MAS, RBI, HKMA, OJK, BNM, and BOT — have all now issued AI governance frameworks requiring explainable, auditable decisioning models in production. MAS’s FEAT principles, developed with the financial industry from 2018 and operationalised through the Veritas framework from 2019, were among the first sectoral AI governance frameworks in the world. RBI’s AI and machine learning governance guidance and HKMA’s AI in banking supervisory guidance have followed. The direction of travel is consistent across every AP jurisdiction: AI systems used in credit, fraud, and AML decisions must be explainable, their decisions must be auditable, and the humans overseeing them must have genuine intervention capability rather than nominal review processes.
The July 2025 MAS enforcement action against nine financial institutions for AML control failures related to the 2023 Singapore case was the clearest signal yet that AP regulators are no longer in a consultation posture. The penalties were not large by global standards — S$27.45 million across nine institutions. What mattered was the specificity of the findings: inadequate alert calibration, weak investigation documentation, and failure to maintain governance infrastructure adequate to the institution’s actual risk profile. These are not findings about fraud outcomes. They are findings about the quality of the decision infrastructure. The signal to every AP bank is direct: having an AML programme that appears adequate on paper is no longer sufficient.
The compliance view and the advantage view produce different institutions
The framing that most AP banks apply to AI governance produces the wrong outcome, and the cost of that framing is now measurable. Under the compliance view: governance is box-ticking for the regulator, documentation is a cost to build after deployment, explainability slows model development, fairness requirements constrain model accuracy, and audit trails are a legal liability requirement. Banks operating with this worldview consistently find that models in production lack the audit trails needed for regulatory examination, that explainability cannot be retrofitted without touching architecture and data pipelines, and that regulators are adversarial rather than cooperative because the evidence of sound process is unavailable.
The advantage view produces materially different outcomes across every dimension that matters.
Banks that treat explainability as a design requirement find that models with built-in explainability receive MAS and RBI sign-off in weeks rather than months. The review process is faster because the examiner can engage with decision logic rather than being handed a model whose outputs cannot be articulated. Banks that treat audit trails as a data asset rather than a compliance obligation find that every logged decision is a labelled training example — the model improves continuously from its own production history in a way that an unlogged system structurally cannot. Banks that treat fairness metrics as a customer acquisition tool rather than a constraint find that documented fairness enables them to serve thin-file segments across Southeast Asia that competitors, constrained by models they cannot defend to regulators, are effectively excluding. This is not a marginal benefit in a region where Indonesia’s underbanked population exceeds 65% of adults and OJK is actively driving financial inclusion requirements.
The financial difference between these two postures is quantifiable. Retrofitting explainability, audit trails, and bias documentation into a production AI model that was not designed to accommodate them typically costs three to five times what it would have cost to build them in from the start. The cost touches data pipelines, model architecture, monitoring infrastructure, and regulatory documentation across the entire deployment. The nine financial institutions penalised by MAS in July 2025 spent years operating under a compliance view. The remediation programmes now required of them will cost far more than early investment in governance infrastructure would have.
The institutions that will compound advantage in AP banking over the next several years are not necessarily those deploying the most technically advanced models. They are those that have built the programme infrastructure to translate model quality into production outcomes at speed, and that have established the audit trail infrastructure to improve continuously from every decision their models see. Governance is not the overhead that constrains the AI programme. It is the mechanism by which model quality compounds.
There are three paths, and the timing determines the magnitude of the advantage
The evidence across AP markets is consistent. The fraud cost multiplier is higher in APAC than anywhere else. Regulatory enforcement is active and specific. Digital-first challengers are widening the decision speed gap with every quarter that incumbents delay. The question facing AP institutions is not whether AI-enabled decisioning matters. It is when to act, and timing now determines the magnitude of the advantage, not just the speed of the benefit.
The first path is to lead. Define the decision portfolio now. Use the four-criteria prioritisation framework to identify the highest-priority decision based on financial consequence, volume, data readiness, and regulatory urgency. Run a properly governed proof of value on that decision, with governance infrastructure — explainability, audit trails, bias monitoring — built in from day one rather than deferred. Use the results to build the board-level business case for enterprise rollout before regulatory pressure forces a reactive programme. Institutions that act in the next six months secure first-mover advantage in model quality that compounds with every decision their system sees. The team that builds the first model builds all subsequent models faster. The regulatory relationship established before enforcement intensifies is cooperative rather than adversarial. And the twelve-month head start in outcome data becomes a structural quality advantage that followers cannot close quickly.
The second path is to follow. Wait for internal consensus, for the technology to mature further, or for early movers to demonstrate the returns before committing. This path avoids early implementation risk and remains viable for some institutions. Its costs are bounded but real: no first-mover advantage in model quality, a regulatory compliance timeline that is shortening as AP enforcement posture tightens, a talent market for AI governance expertise that becomes more competitive as demand concentrates, and competitor models that will have twelve to eighteen months of outcome data — a quality gap that is real and compounds.
The third path is to defer. Continue with current rule-based and batch-scoring systems and revisit the question when external pressure forces action. In practice, this path ends with a crisis-driven reactive programme — the worst possible conditions under which to build AI infrastructure that is supposed to be reliable, explainable, and fair. The structural cost disadvantage compounds annually. Customer attrition to faster competitors accelerates with every credit decision that takes days rather than seconds. Regulatory risk accumulates as models without governance face examination findings. The talent and technology markets become less favourable each quarter. And when the institution eventually acts under crisis conditions, it does so with a compressed timeline, an adversarial regulatory relationship, and none of the outcome data that early movers have been accumulating for years.
The question is not whether to build the advantage. It is whether to build it deliberately, or inherit the disadvantage by default.
Part 3 of 3.
Sources
Monetary Authority of Singapore. FEAT Principles: Fairness, Ethics, Accountability and Transparency in the Use of AI and Data Analytics in Singapore’s Financial Sector. 2018. Monetary Authority of Singapore. Veritas Framework. 2019; updated 2023. Monetary Authority of Singapore. MAS Imposes Financial Penalties on Financial Institutions for Anti-Money Laundering Control Breaches. July 2025. Reserve Bank of India. AI and Machine Learning Governance Guidance. RBI, ongoing. Hong Kong Monetary Authority. AI in Banking Supervisory Guidance. HKMA, 2024. Fenergo. Half-Year Financial Institution Enforcement Report, H1 2024. August 2024.
