Model AI Governance Framework
Asia’s first AI governance framework. Four key areas, two guiding principles, technology-agnostic by design.
Building with GenAI or LLMs? The 2024 GenAI Framework addresses your specific risks.
Go to GenAI & Agentic AI Frameworks →The Two Guiding Principles
Every provision in the framework traces back to these two foundational commitments. They are the “why” behind the “what.”
Explainable, Transparent and Fair
Decisions made by AI should be explainable, transparent, and fair. Organizations should strive to ensure their AI use reflects these objectives throughout the system lifecycle, from data collection through model deployment and monitoring.
Human-Centric
AI solutions should be human-centric. The protection of human interests, well-being, and safety should be the primary considerations in AI design, development, and deployment. Harm prevention takes precedence over performance optimization.
The Four Key Areas
Structured governance across internal controls, human oversight, operational rigor, and stakeholder communication.
- Apply a risk-based approach: assess probability and severity of harm to determine the level of human involvement needed.
- Human-in-the-loop: AI recommends, human makes the final decision. Used for high-risk, high-impact scenarios.
- Human-over-the-loop: Human supervises and can intervene or take over at any point. Appropriate for medium-risk applications.
- Human-out-of-the-loop: AI acts autonomously with periodic review. Reserved for low-risk, well-validated systems.
- Define and document the organization’s risk appetite before selecting the oversight model for each AI use case.
- Data management: Establish data lineage, enforce data quality standards, and minimize bias in training and operational data.
- Algorithm selection: Document the rationale behind model and algorithm choices, including tradeoffs between accuracy and explainability.
- Ensure explainability, repeatability, reproducibility, and traceability of AI outputs and decisions.
- Conduct regular model tuning and active monitoring after deployment to detect drift, degradation, and emerging risks.
- Treat the AI lifecycle as a continuous learning process, not a one-directional pipeline. Feedback loops from monitoring should inform future development.
- Provide general disclosure when AI is being used to make or support decisions that affect individuals.
- Inform users explicitly when they are interacting with an AI system rather than a human.
- Use counterfactual explanations and comparisons to help users understand how AI-driven decisions were reached.
- Offer opt-out options where feasible, allowing individuals to request human review of AI-assisted decisions.
- Establish accessible feedback channels and formal appeal mechanisms so affected parties can raise concerns.
Design Philosophy
Three design decisions that make the framework broadly applicable and adoption-friendly.
Technology-Agnostic
Does not focus on specific systems, software, or technology. The framework applies regardless of development language, data storage method, or deployment environment. It governs the outcomes, not the stack.
Sector-Agnostic
Provides baseline considerations for any industry. Specific sectors such as finance (MAS FEAT), healthcare (AIHGle), or legal (MinLaw Guide) can layer additional requirements on top without conflict.
Voluntary Adoption
Not legally binding, but carries significant regulatory weight through institutional backing from the Infocomm Media Development Authority (IMDA), Personal Data Protection Commission (PDPC), and international recognition. Voluntary does not mean optional in practice for regulated industries.
ISAGO: Self-Assessment Companion
The practical implementation guide that turns the Model Framework into actionable self-assessment.
Implementation and Self-Assessment Guide for Organisations
The Implementation and Self-Assessment Guide for Organisations (ISAGO) was co-developed by the Personal Data Protection Commission (PDPC) and the World Economic Forum Centre for the Fourth Industrial Revolution. It translates the Model Framework’s four key areas into a structured set of self-assessment questions that organizations can use to evaluate their AI governance maturity. Organizations looking to move from self-assessment to automated testing should explore AI Verify, which tests against 11 governance principles derived from this framework.
- Checklist of self-assessment questions mapped to each of the four key areas
- Extensive real-world industry examples and practices contributed by over 60 organizations
- Gap identification: helps organizations pinpoint exactly where their governance falls short
- Practical remediation guidance for each identified gap
- Designed for iterative use as AI governance programs mature over time
Co-Developed By
PDPC – Personal Data Protection Commission
WEF C4IR – World Economic Forum Centre for the Fourth Industrial Revolution
Framework Evolution
How the Model Framework grew into a family of governance instruments covering traditional AI, generative AI, and autonomous agents.
Related Tools
Downloadable templates to accelerate your implementation of the Model Framework.
Model Framework Self-Assessment Checklist
Map your AI governance posture against all four key areas of the Model AI Governance Framework. Includes scoring criteria and gap indicators.
GenAI and Agentic AI Governance Setup Guide
Step-by-step implementation guide covering the 9 GenAI trust dimensions and 4 agentic governance areas. Built for organizations scaling beyond traditional AI.