Zuletzt aktualisiert am:
The MAS FEAT Principles are four principles covering Fairness, Ethics, Accountability and Transparency, published by the Monetary Authority of Singapore on November 12, 2018, to guide the responsible use of artificial intelligence and data analytics in Singapore's financial sector. They are non-binding guidance, not enforceable rules, and are supported by the Veritas assessment methodology.
What does FEAT stand for?
The FEAT Principles set out four areas of expectation for financial institutions using AI and data analytics (AIDA):
Fairness. AIDA-driven decisions should not systematically disadvantage individuals or groups without justification. Decisions should be justifiable, data and models should be reviewed regularly for accuracy and unintended bias, and the attributes used should be relevant to the decision.
Ethics. AIDA use should align with the institution's own ethical standards, values and codes of conduct, and be held to at least the same standard as decisions made by people.
Accountability. Internal accountability for AIDA-driven decisions, approved by an appropriate authority, and external accountability to the customers those decisions affect.
Transparency. Proactive disclosure to data subjects about AIDA use, and clear explanation of AIDA-driven decisions on request.
Are the FEAT Principles mandatory?
No. FEAT is non-binding guidance issued by MAS, not a regulation carrying direct penalties. It functions as a supervisory expectation. MAS uses it as the reference point for how Singapore financial institutions should govern AI, and firms are expected to explain their approach against it. It works more like a standard than a rule. Firms are not fined for non-adherence, but they are expected to hold a defensible position.
How do firms demonstrate FEAT adherence?
MAS supplemented the principles with the Veritas Initiative, a consortium effort that produced assessment methodologies translating each FEAT principle into testable steps, along with open-source toolkits. Singapore's AI Verify framework offers a further testing route. Together these move FEAT from stated principle to documented evidence, which is what a supervisor asks to see.
What does FEAT mean for AI governance programs?
FEAT places weight on the first line of defense. The business owner deploying a model is accountable for its outcomes, and not only the model risk or compliance function. That makes intake and ownership records as important as validation. Most firms operationalize it by recording, for every AIDA use case, a named accountable owner, the fairness assessment performed, the data attributes used, and the customer-facing disclosure. This is the same evidence base needed for automated AI governance.
Praxisbeispiel:
A Singapore bank assesses its AI mortgage advisory model against the Fairness principle using the Veritas methodology. The review finds that the model uses postal code as a feature, which correlates with ethnicity in some districts and acts as a proxy variable. The bank removes the feature, re-tests for disparate outcomes across applicant groups, records the justification for every remaining attribute, and assigns accountability for the model's outcomes to the named head of the mortgage business instead of the data science team that built it.
