Singapore - AI Fairness Principles (2018)

Overview

The Monetary Authority of Singapore (MAS) published the Principles to Promote Fairness, Ethics, Accountability and Transparency (FEAT) in the Use of Artificial Intelligence and Data Analytics in Singapore’s Financial Sector on 12 November 2018. FEAT provides high‑level, non‑binding principles for firms providing financial products and services to consider when adopting and using AIDA (artificial intelligence and data analytics). The document aims to build public confidence in AIDA by encouraging firms to embed governance, risk management and ethical considerations across the model lifecycle. MAS developed FEAT alongside industry partners and other Singapore agencies; the full text and supporting materials can be found in the official MAS publication and subsequent MAS initiatives such as the MAS Veritas initiative and the MAS information papers on implementation. FEAT is intended to be contextually applied and scaled according to the materiality and potential impact of AIDA use cases.

Definitions

FEAT defines key constructs used throughout the guidance. "AIDA" refers to artificial intelligence and data analytics tools used in decision‑making for financial products and services. "Fairness" addresses outcomes across individuals or groups, focusing on avoiding systematic disadvantage unless justified and documented. "Ethics" requires alignment with firm values, codes of conduct and societal norms, including consideration of dignity and non‑discrimination. "Accountability" requires explicit governance structures, roles and responsibilities for AIDA projects (e.g., model owners, independent validators, board oversight). "Transparency" emphasizes meaningful disclosure, explainability and sufficient documentation to enable stakeholders — including customers and supervisors — to understand material decisions. FEAT encourages firms to interpret and operationalise these definitions in light of local law, data protection requirements and sectoral norms.

Governance and Institutional Framework

FEAT places governance at the centre of responsible AIDA use. Boards and senior management are expected to own AIDA governance, set risk appetite and allocate resources for oversight. Firms should establish AIDA policies, an inventory of material AI systems, and clear role‑based responsibilities (model owners, compliance, risk, internal audit, validation teams). MAS recommends independent review or validation of material models and escalation channels for model issues. Where group structures exist, MAS expects Singapore‑facing governance representation and that group‑level committees account for local risks and regulatory obligations. FEAT also encourages cross‑functional teams (data scientists, legal, compliance, business units, user experience) and training to ensure managerial and technical competency. For examples, MAS points to industry efforts like the Veritas consortium which produced operational assessment methodologies aligned to FEAT.

Key Focus Areas

FEAT organises practical attention across the model lifecycle. Data governance: ensure data quality, provenance, representativeness and lawful use, including PDPC obligations on personal data. Model development: select appropriate algorithms, features and training regimes that mitigate bias; document design choices. Testing and validation: perform pre‑deployment testing for accuracy, fairness, robustness and resilience to distributional shifts; validate performance across relevant subgroups. Explainability and transparency: provide meaningful explanations to affected customers where appropriate and maintain documentation and model cards for supervisors and internal stakeholders. Human oversight and decisioning: ensure human review in high‑impact decisions and set clear intervention points and escalation procedures. Monitoring and incident management: implement continuous monitoring for performance drift, emergent bias and operational issues; maintain logs and traceability. Vendor and third‑party management: ensure contractual and oversight safeguards where models or data are sourced externally. FEAT also emphasises proportionality — the intensity of governance should match the potential impact of the AIDA application.

Implementation Framework

FEAT is designed to be operationalised by firms through governance controls and lifecycle integration. MAS recommends firms map their use cases, classify materiality, and then apply layered controls: policy and standards, model inventories, project intake and review procedures, risk assessments, independent validation, deployment gating, post‑deployment monitoring and consumer communications. Firms are encouraged to document decisions and control rationales in a central repository and to build capabilities including technical skills, legal/risk expertise and ethical review processes. MAS has worked with industry to produce the Veritas assessment methodologies and toolkits to help organisations translate FEAT into measurable controls and to generate assessment reports that can be used for internal governance and supervisory engagement. Further guidance on implementation was later published by MAS in thematic reviews and information papers covering fairness implementation observations.

Monitoring and Evaluation

Monitoring is critical to FEAT’s objectives. MAS expects firms to implement ongoing monitoring of model performance, fairness metrics and operational health metrics; set thresholds and triggers for re‑training or human intervention; and maintain audit trails and logs for key decisions and model versions. Monitoring should also include periodic review of data inputs to detect shifts in distribution or representativeness, and regular validation of outcomes against intended objectives. MAS’s thematic reviews emphasise that many firms were in early stages of setting up such monitoring, and MAS provides recommendations and examples of good practice. Veritas tools and MAS information papers offer methodological guidance on measurement and reporting of fairness and related metrics.

Penalties, Liability, and Appeals

FEAT itself is non‑binding guidance; it does not create new statutory penalties. However, MAS indicates that ineffective AIDA governance that results in consumer harm, breaches of licensing conditions, or contraventions of existing laws (e.g., consumer protection, anti‑money laundering, or data protection obligations enforced by the Personal Data Protection Commission) may lead to supervisory actions, enforcement measures or civil liability under applicable statutes or contract law. MAS has used thematic reviews and supervisory engagement to encourage remediation and has articulated that poor governance could attract formal supervisory responses in severe cases. Affected consumers retain usual civil and administrative routes for redress; firms are encouraged to maintain complaint handling and redress mechanisms for AIDA‑driven decisions.

Relationship to Other Instruments

FEAT is explicitly linked to other Singapore AI governance instruments. MAS developed FEAT in consultation with the Personal Data Protection Commission (PDPC) and the Infocomm Media Development Authority (IMDA) to align with the PDPC’s Model AI Governance Framework and IMDA’s advisory activities. FEAT operates alongside Singapore financial sector regulations (banking, payments, insurance licensing regimes) and complements MAS supervisory expectations on operational resilience, model risk management and consumer protection. Internationally, FEAT aligns conceptually with OECD AI principles, EU ethical AI approaches and other central bank/regulator initiatives; MAS has emphasised cross‑jurisdictional relevance through the Veritas consortium and multi‑lateral dialogues.

International Alignment

FEAT is widely cited as an early regulator‑led articulation of AI ethics in finance and resonates with international efforts to promote trustworthy AI. The four FEAT pillars map to common international themes — fairness/non‑discrimination, human‑centric ethics, accountable governance and transparency/explainability — used by OECD, EU, and national AI strategies. MAS stresses interoperability and comparability, recommending that firms align FEAT implementation with other jurisdictional requirements where they operate. Initiatives such as Veritas and MAS’s public materials (including implementation observations) are shared to facilitate cross‑border learning and standardisation of assessment approaches.

Implementation Timeline

Sources and References