Blog / Governance

Singapore publishes a governance framework for agentic AI

Singapore presents its text as the first comprehensive guide to deploying agents responsibly. It sets a clear rule: humans remain ultimately accountable.

By Vincent Dorange, , 2 min read

22 Jan 2026

Singapore presents its agentic AI governance framework in Davos

On 22 January 2026, at the World Economic Forum in Davos, Singapore's Minister for Digital Development and Information, Josephine Teo, announced the Model AI Governance Framework for Agentic AI 1. Developed by IMDA, the country's digital authority, it builds on the AI governance framework Singapore introduced in 2020. The announcement presents it as the first in the world to give enterprises a comprehensive guide to deploying agents responsibly.

Why agents need their own framework

The text starts from a simple difference. Unlike traditional or generative AI, an agent reasons and takes actions on behalf of its user. It may have access to sensitive data and be able to change its environment, for instance by updating a customer database or making a payment, which opens the door to unauthorised or erroneous actions. IMDA also flags automation bias, the tendency to over-trust a system that has worked reliably so far.

Four dimensions

The framework addresses organisations that build their own agents as well as those that buy third-party solutions. It recommends technical and non-technical measures across 4 dimensions 1:

  • assess and bound risks upfront, by choosing suitable use cases and limiting the agent's powers, both its autonomy and its access to tools and data;
  • make humans meaningfully accountable, by defining the checkpoints at which their approval is required;
  • implement technical controls across the lifecycle, such as baseline testing and access limited to whitelisted services;
  • enable end-user responsibility through transparency and training.

IMDA describes it as a living document, open to feedback and case studies, and announces guidelines for testing agentic applications.

What the framework leaves to each organisation

The second dimension is the most demanding. Defining the checkpoints where a human approves means knowing, for each of the agent's decisions, what it commits and what it would cost if it were wrong. The framework sets the requirement; it leaves each company to draw the line.

That is precisely the work the Agentic Commerce Framework® proposes: write each agent's mandate, place the human according to each decision's criticality and name the person who answers for it. Singapore and the ACF share a starting point that fits in one sentence: an agent acts, a person answers.

Sources

Links to the official texts and original publications.

  1. 1Singapore Launches New Model AI Governance Framework for Agentic AI, Infocomm Media Development Authority (IMDA), 22 January 2026.

Glossary terms used in this article

AI agent mandateDecision criticality

Vincent Dorange, creator of the Agentic Commerce Framework®

Also worth reading

95%

of organisations surveyed faced at least one agent-related incident in 12 months

Governance

95% of organisations have already had an AI agent incident, says Gartner

25

organisations interviewed; none lets its agents act with unrestricted autonomy

Governance

AI agents at work: what the OECD learned from 25 organisations

4 levers

transaction caps, budgets, approved counterparties and escalation rules, per Circle

Agentic economy

Agentic economy: agents that spend within delegated permissions

All articles