MAS与新加坡金融服务AI：FEAT原则、Veritas和新指南
MAS拥有亚洲最发达的金融AI治理框架。FEAT原则、Veritas方法论，以及2025年11月发布的新AI风险管理指南。
Key Takeaways
MAS's FEAT principles (Fairness, Ethics, Accountability, Transparency) are the primary framework for AI governance in Singapore financial services. All significant customer-facing AI should be assessed against FEAT.
The Veritas initiative, developed by MAS with industry partners, provides detailed methodology for implementing FEAT assessments — particularly fairness testing for credit, insurance, and HR AI.
MAS's model risk management expectations, articulated in its guidelines for banks and insurers, apply to AI and ML models — not just traditional statistical models.
Singapore financial institutions with EU customers must comply with both MAS requirements and the EU AI Act's extraterritorial obligations — the frameworks are largely compatible but require careful mapping.
MAS has signalled increasing supervisory focus on AI risk governance in its annual supervisory priorities — AI governance is now a standing examination item for major financial institutions.
"仅供参考。本文不构成法律、监管、财务或专业建议。如需具体指导，请咨询合格专家。"
MAS as the APAC benchmark for financial AI governance
The Monetary Authority of Singapore has emerged as the leading regulatory voice on AI governance in Asian financial services. In November 2025, MAS went beyond its existing FEAT principles by issuing a consultation paper proposing formal Guidelines on AI Risk Management — a more structured set of supervisory expectations that, once finalised, will apply to all MAS-regulated financial institutions.
The proposed guidelines cover four areas: oversight of AI risk management (board and senior management accountability, governance structures, risk culture); key AI risk management systems, policies and procedures (AI inventory, materiality assessments, risk registers); AI lifecycle controls (data management, fairness and transparency, human oversight, third-party risk management); and capabilities and capacity for AI risk management (skills, tools, expertise). The consultation closed in January 2026 and final guidelines are expected to be published in 2026.
The FEAT principles and Veritas: still the operational framework
While the new guidelines represent an evolution, MAS's existing FEAT principles (Fairness, Ethics, Accountability, Transparency) remain the foundational framework for AI governance in Singapore financial services. FEAT is operationalised through the Veritas initiative, which provides specific quantitative methodology for assessing fairness in credit, fraud detection, and customer marketing AI.
Fairness: AI systems should not produce discriminatory outcomes based on characteristics that should not influence financial decisions. Veritas provides specific metrics for measuring fairness across demographic groups and use cases. Proxy discrimination — where a model uses variables that correlate with protected characteristics to produce disparate outcomes — is within scope even when the protected characteristics are not explicitly used.
Ethics: AI use must be consistent with the values and ethical standards expected of financial institutions. MAS has been explicit that this is not aspirational — it must be operationalised through governance structures, training, and incentive systems.
Accountability: Clear accountability for AI outcomes must be established. For each significant AI system, there should be a named owner responsible for governance, performance monitoring, and escalation. Diffused accountability is not accountability.
Transparency: AI systems should be explainable to regulators, internal auditors, and affected customers at appropriate levels of technical depth. Both model transparency (internal explainability) and customer transparency (meaningful explanation of AI-influenced decisions) are expected.
The 2026 regulatory evolution
In January 2026, IMDA released the Model AI Governance Framework for Agentic AI — extending Singapore's voluntary governance framework to cover autonomous AI agents capable of independent decision-making. This makes Singapore one of the first jurisdictions globally to provide structured governance guidance for agentic AI systems. Financial institutions deploying AI agents for customer service, investment management, or compliance monitoring should assess these systems against the new framework.
MAS is also finalising its formal AI Risk Management Guidelines following the 2025 consultation. These guidelines, once published, will represent a significant step toward mandatory AI governance expectations for Singapore's financial sector — moving beyond the voluntary FEAT principles toward supervisory expectations with clear compliance implications.