AxiLayer AI · Frameworks

Singapore Model AI Governance Framework for Agentic AI

Four dimensions for agents that act, issued January 2026, mapped to the controls that bound autonomy and keep a person accountable.

Run the Agent Exposure Check See the mapping

Last verified: 30 October 2026

The framework

The first government framework written for agents

What it is

The Model AI Governance Framework for Agentic AI, issued by Singapore's Infocomm Media Development Authority on 22 January 2026 at the World Economic Forum. Four dimensions: assess and bound risks upfront, make humans meaningfully accountable, implement technical controls and processes, enable end-user responsibility.

Why it matters beyond Singapore

It is the first government framework written for agents rather than models, it is referenced by MAS's AI Risk Management Guidelines of 7 October 2026, and buyers across ASEAN, the Gulf and Europe use it as a checklist.

How AxiLayer uses it

The four dimensions map onto the twelve phases almost one to one; the evidence the passport carries answers each dimension with a record, not a statement.

Mapping

Each requirement, the control that meets it

Every row is a control that is live in AxiSentinel or AxiReady today, with the phase it belongs to.

Requirement	What AxiSentinel and AxiReady do	Phases
Assess and bound risks upfront	Classify and plan per agent: risk tier, allowed tools, data scope and environment; AxiReady readiness score before go-live.	01, 02
Bound autonomy through tool access and permissions	MCP Gateway approves every tool; named identity and role scope; unapproved tools blocked.	03, 04
Make humans meaningfully accountable	A named owner per agent in the inventory; high-impact actions wait for a named person; the confirmation and reason are recorded.	04, 06
Human oversight that can override, intercept or review	Authority gates intercept; approvals inbox reviews; quarantine and kill switch override on one confirmed decision.	05, 06, 08
Implement technical controls and processes	Tool guardrails at the gateway, least-privilege scope enforced per agent, LLM proxy policy on every model call.	03, 04, 05
Test before deployment	Agentic test suite: task execution, policy compliance and tool-use accuracy, plus the OWASP cases.	10
Staged rollout and real-time monitoring	On Demand first, Always On after; behavior guard, AI radar and monitoring plans with Article 72 style data checks.	07, 09
Enable end-user responsibility	Disclosure and escalation evidence in the passport; Academy training for the people who approve and stop.	11, 12

Sources: IMDA, Model AI Governance Framework for Agentic AI, 22 January 2026; MAS AI Risk Management Guidelines, 7 October 2026; Baker McKenzie client note, January 2026.

Evidence

What the Compliance Passport carries for this framework

Bounded-autonomy record

For each agent: tools approved, scope, owner, and every change to them, signed.

Oversight log

Every hold, release, quarantine and stop with the person who confirmed it.

Test and rollout evidence

Pre-deployment test report and the staged-rollout monitoring record.

Three layers, twelve phases

Twelve phases. One platform. One signature.

Phase 01FindDiscover and score

Phase 02FindClassify and plan

Phase 03ControlApprove every tool

Phase 04ControlName every agent

Phase 05ControlGate every action

Phase 06ControlHold for a person

Phase 07ControlWatch behavior

Phase 08ControlQuarantine and stop

Phase 09ControlMonitor to the article

Phase 10ProveTest and re-test

Phase 11ProveSign the evidence

Phase 12ProveVerify, free, forever

AxiReady finds. AxiSentinel gates, holds, quarantines and stops, Always On or On Demand. A qualified AxiLayer AI auditor signs against Singapore Agentic AI.

Singapore Model AI Governance Framework for Agentic AI

The first government framework written for agents

Each requirement, the control that meets it

What the Compliance Passport carries for this framework

See where your agents stand in three minutes