Singapore's Model AI Governance Framework, second edition, was released by the Personal Data Protection Commission with IMDA on 21 January 2020 as a voluntary guide for organisations deploying AI at scale. It covers four areas: internal governance structures and measures, determining the level of human involvement in AI-augmented decision-making (human-in-the-loop, human-out-of-the-loop and human-over-the-loop), operations management including data and model management, and stakeholder interaction and communication. For banks it sits beneath MAS's sector guidance: paragraph 2.13 points finance firms to MAS's FEAT Principles and says adopting the framework 'does not mean that organisations are in compliance with such sector-specific laws'.
OFFICIAL TEXT: imda.gov.sg ↗ · IN FORCE · PDPC SINGAPORE
| Document | Model AI Governance Framework (Second Edition) — Model Artificial Intelligence Governance Framework, Second Edition |
| Issued by | Personal Data Protection Commission (Singapore) — administers and enforces the Personal Data Protection Act 2012 |
| Type | Framework |
| Status | In force |
| Published | Jan 21, 2020 |
| Applies to | Private-sector organisations in any sector deploying AI solutions at scale, including banks; voluntary and accountability-based, issued by the PDPC with IMDA. It does not replace sector rules: banks remain subject to MAS requirements, which the framework itself points to |
| Official source | imda.gov.sg ↗ |
| Use cases | AI governance (general) · Model risk management · Data & privacy · Credit scoring & underwriting · Customer-facing chatbots |
What are the key points of Model AI Governance Framework (Second Edition)?
- Four key areas (paragraph 3.1): internal governance structures and measures; determining the level of human involvement in AI-augmented decision-making; operations management; and stakeholder interaction and communication.
- Three degrees of human oversight: human-in-the-loop, human-out-of-the-loop and human-over-the-loop (or human-on-the-loop), chosen through a risk-based methodology weighing the probability and severity of harm.
- The second edition 'includes additional considerations (such as robustness and reproducibility) and refines the original framework', and clarified the human-over-the-loop approach; the first edition was released on 23 January 2019.
- Voluntary: adopting it 'will not absolve organisations from compliance with current laws and regulations', but can help demonstrate accountability-based data practices under the PDPA (paragraph 2.12).
- Paragraph 2.13 names MAS's FEAT Principles as sector guidance for financial firms; MAS in turn revised FEAT paragraph 1.4 in February 2019 to align with the framework.
- Paragraph 3.47 suggests disclosing how an AI decision may affect individuals, using the example of telling individuals that credit ratings may lead to a loan refusal and that the decision is reversible if they provide more evidence of creditworthiness.
- It is not intended for organisations merely deploying updated off-the-shelf software that happens to incorporate AI (paragraph 2.2); its guidance was later 'incorporated into and expanded under' the Trusted Development and Deployment dimension of the 2024 Model AI Governance Framework for Generative AI.
What did Model AI Governance Framework (Second Edition) change for banks?
The framework is the root of Singapore's AI governance stack: the IMDA frameworks for generative AI (2024) and agentic AI (2026), the AI Verify testing framework and the PDPC's 2024 advisory guidelines all build on it or refer back to it, and MAS's FEAT Principles were aligned with it. Banks rarely cite it directly, but its human-oversight taxonomy is the vocabulary MAS and IMDA documents still use.
Does the Model AI Governance Framework apply to banks?
It is voluntary guidance for all sectors, so banks can use it, but paragraph 2.13 says regulated sectors such as finance remain subject to sector-specific rules and guidance, naming MAS's FEAT Principles, and that adopting the framework does not mean compliance with them.
Who issued Singapore's Model AI Governance Framework?
The Personal Data Protection Commission, with the Infocomm Media Development Authority. The first edition was released on 23 January 2019 and the second edition on 21 January 2020.
What does human-over-the-loop mean in the Model AI Governance Framework?
A human monitors or supervises the AI system and can take control or adjust parameters when it encounters unexpected or undesirable events, as distinct from human-in-the-loop (a human makes or approves each decision) and human-out-of-the-loop (no human oversight of execution).
| Date | Document | Status |
|---|---|---|
| Jul 20, 2026 | PDPC Advisory Guidelines on Personal Data in Generative AI — Advisory Guidelines on Use of Personal Data in Generative AI | In force |
| Mar 1, 2024 | PDPC Advisory Guidelines on AI Recommendation and Decision Systems — Advisory Guidelines on Use of Personal Data in AI Recommendation and Decision Systems | In force |
| Oct 15, 2026 | FSB AI sound practices (final report, Oct 2026) — Sound Practices for Responsible Adoption of Artificial Intelligence (AI): Final report | Final |
| Oct 15, 2026 | HKMA "ART" Framework / AI supervisory guidance — "ART" Framework and Updated Supervisory Guidance for Responsible Adoption of Artificial Intelligence | Final |
| Oct 7, 2026 | MAS Guidelines on Artificial Intelligence Risk Management — Guidelines on Artificial Intelligence Risk Management for Financial Institutions | Final |
| Sep 30, 2026 | SB 947 — Employment: Automated Decision Systems (No Robo Bosses Act) | Final |
Follow every document these regulators publish
when one of these regulators moves, the next morning's brief says so · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning