What does MAS FEAT Principles say about AI in banking?

What are the key points of MAS FEAT Principles?

• Fourteen principles in four groups: Fairness (principles 1-4: justifiability, and accuracy and bias), Ethics (5-6), Accountability (7-11: internal and external) and Transparency (12-14).
• Fairness: individuals or groups are not systematically disadvantaged through AIDA-driven decisions unless justified (principle 1), use of personal attributes as input factors is justified (2), and data and models are regularly reviewed and validated for accuracy, relevance and unintentional bias (3-4).
• Ethics: AIDA use is aligned with the firm's ethical standards, values and codes of conduct (5) and AIDA-driven decisions are held to at least the same ethical standards as human-driven decisions (6).
• Accountability: use of AIDA is approved by an appropriate internal authority (7), firms are accountable for both internally developed and externally sourced models (8), and management and the Board are proactively made aware of AIDA use (9); data subjects get channels to enquire about, appeal and request review of decisions (10), and verified supplementary data they provide is taken into account (11).
• Transparency: use of AIDA is proactively disclosed to data subjects as part of general communication (12), and on request data subjects receive clear explanations of what data is used, how it affects the decision (13) and the consequences for them (14); paragraph 8.2 says explanations need not expose intellectual property or source code.
• Paragraph 2.3 lets firms calibrate actions to materiality, listing indicative considerations such as the extent of AIDA use, model complexity, extent of automation, severity and probability of impact on individuals, monetary and regulatory impact, and options for recourse.
• Paragraph 3.2 defines AIDA as 'artificial intelligence or data analytics', defined as technologies that assist or replace human decision-making, so the Principles reach straightforward analytics as well as complex models.
• Paragraph 1.4 was revised on 7 February 2019 to align with the Personal Data Protection Commission's Model AI Governance Framework.

What did MAS FEAT Principles change for banks?

FEAT was one of the first sector-specific AI principles issued by a financial regulator. It gave Singapore firms a shared vocabulary for responsible AI before generative AI existed, and MAS has since layered tools and supervisory papers on top of it (the Veritas Initiative, Project MindForge, the December 2024 AI model risk management information paper and the 2025 draft AI risk management Guidelines). For banks it remains the ethical and customer-outcome baseline, while the newer MAS documents supply the risk-management expectations.

What do the MAS FEAT Principles expect of banks that use AI?

The MAS FEAT Principles set 14 voluntary principles for firms that use artificial intelligence or data analytics in decisions about financial products and services: AIDA-driven decisions should be justified and free of unintended bias, aligned with the firm's ethical standards, approved by an appropriate internal authority, open to enquiry and appeal by affected individuals, and explained on request. The Principles are not prescriptive and do not replace existing governance frameworks or legal requirements, and firms may scale the controls to the materiality of each decision using the considerations in paragraph 2.3. A bank using AI for credit decisions, segmentation or customer interaction can use them as a checklist for model governance, customer recourse and disclosure. MAS describes them as foundational and has built later guidance, including its draft AI risk management Guidelines, on top of them.

FEAT is principles-based and illustration-led: each group of principles is followed by illustrative examples (for instance, justifying each factor used in a credit-card limit increase, or telling a caller they are speaking to an AI phone operator) that MAS describes as purely illustrative. It sits alongside, not in place of, firms' internal governance and the Personal Data Protection Commission's Model AI Governance Framework, with which MAS aligned paragraph 1.4 in February 2019.

MAS has treated FEAT as the first layer of a stack. The Veritas Initiative, started with an industry consortium in November 2019, produced assessment methodologies, a toolkit and case studies; Project MindForge extended the work to generative AI; and the November 2025 consultation on Guidelines on AI Risk Management states that the Guidelines build on FEAT, experience with industry and MAS's information papers. FEAT therefore answers the 'what outcomes should AI decisions deliver' question, and the newer documents answer 'what risk management supports it'.

WHAT THIS MEANS IN PRACTICE

• Map the 14 principles to existing model governance, conduct and complaints procedures rather than building a separate FEAT process.
• Document the justification for personal attributes and other sensitive inputs in credit, pricing and segmentation models.
• Make sure customers affected by an AI-driven decision have a route to enquire, appeal and have supplementary data considered.
• Disclose AI use to customers in general communications and be ready to explain the data used and the effect on the decision without disclosing source code.
• Treat the Principles as the fairness and ethics baseline when applying the MAS risk-management expectations in the AI model risk management information paper and the draft Guidelines.