Singapore’s Model AI Governance Framework: From Traditional AI to Agentic AI

Singapore has firmly established itself as Asia’s leading voice in artificial intelligence governance. Through the Infocomm Media Development Authority (IMDA) and the AI Verify Foundation, the nation has methodically evolved its flagship guidance to match every wave of technological disruption.

What started in 2020 as a framework for traditional machine learning has now expanded into the world's first comprehensive governance standard for Agentic AI, systems capable of autonomous reasoning, planning, and multi-step execution. For startups, enterprise leaders, and aspiring compliance consultants navigating the ISO 42001 / AIMS landscape, understanding this three-generation evolution is essential for building trust and ensuring safe deployment.

The Three Generations of Singapore's AI Governance

Singapore's regulatory strategy is built on a voluntary, principles-based model rather than heavy-handed statutory mandates. This flexibility has made it the gold standard across Southeast Asia, heavily influencing regional frameworks and aligning closely with global standards like the EU AI Act.

1. The Original Framework (2020): Traditional AI

Published by the Personal Data Protection Commission (PDPC) and IMDA, the inaugural framework focused on recommendation engines, classification models, and predictive analytics. It established two core guiding principles:

• AI decision-making must be explainable, transparent, and fair.

• All AI solutions must remain human-centric.

It also organized responsibilities into four key pillars: internal governance structures, human oversight models, operations management (across the data lifecycle), and stakeholder communication.

2. The GenAI Framework (January 2024): Generative AI

Developed in collaboration with over 70 organizations, the second-generation framework extended governance to large language models (LLMs) and generative tools. It introduced nine critical focus areas, including:

• Accountability across the development-to-deployment chain.

• Data governance, privacy, and copyright management.

• Trusted development, red-teaming, and rigorous safety testing.

• Content provenance, such as watermarking and metadata tracking to identify AI-generated media.

• Cybersecurity defenses against prompt injection and data poisoning.

3. The Agentic AI Framework (January 2026): Autonomous Agents

Unveiled at the World Economic Forum in Davos, Singapore introduced the world's first governance framework specifically designed for agentic AI.

Unlike passive chatbots or static models, agentic systems set and pursue goals autonomously, interact with external tools, and execute workflows without constant human prompts. This introduces unique governance challenges:

• Cascading Actions: Autonomous agents can trigger complex, multi-step chains of actions across software systems, making outcomes harder to predict.

• Multi-Agent Coordination: Interactions between multiple independent AI agents can create emergent behaviors that are difficult to govern at an individual level.

• Accountability Gaps: Determining responsibility when an autonomous agent makes a decision no human directly authorized.

To combat these risks, the 2026 framework mandates explicit safety boundaries, real-time human override mechanisms, and comprehensive audit logging for all agent-driven actions.

Practical Tooling: AI Verify, Project Moonshot, and ISAGO

Singapore’s approach moves far beyond theory by offering concrete, open-source technical tools that organizations can implement immediately:

• AI Verify: A government-developed, open-source testing framework maintained by the AI Verify Foundation (now comprising over 90 member organizations). It allows companies to run standardized technical tests for fairness, transparency, and robustness.

• Project Moonshot: The world's first open-source evaluation toolkit designed specifically for LLMs, offering robust red-teaming and baseline safety checks.

• ISAGO (Implementation and Self-Assessment Guide for Organizations): A practical companion tool. ISAGO 2.0 bridges organizational self-assessment questionnaires with AI Verify’s technical testing into a unified compliance workflow.

A Phased Implementation Roadmap for Organizations

Adopting Singapore's framework shouldn't be approached all at once. Organizations can build maturity incrementally through a four-phase roadmap:

1. Phase 1 (Foundation): Assign AI governance ownership within your risk or compliance team. Complete an ISAGO self-assessment to establish your current baseline maturity.

2. Phase 2 (Traditional AI Governance): Implement the four pillars from the 2020 framework and run AI Verify tests on core legacy models.

3. Phase 3 (GenAI Governance): Map your generative tool usage against the nine 2024 focus areas, implement content provenance tracking, and execute red-teaming via Project Moonshot.

4. Phase 4 (Agentic AI Governance): For systems deploying autonomous agents, define strict safety boundaries, deploy real-time monitoring for cascading actions, and ensure foolproof audit logging.