Model AI Governance Framework for Agentic AI

IMDA's Model AI Governance Framework for Agentic AI is Singapore's governance guidance for AI agents. The version in the source text is version 1.5, published on 20 May 2026. It is organised around four dimensions, which the document gives as:

1. Assess and bound the risks upfront.
2. Make humans meaningfully accountable.
3. Implement technical controls and processes.
4. Enable end-user responsibility.

The document describes these four dimensions as an iterative process. If an anomaly is found during implementation or monitoring, the earlier dimensions should be revisited. The text does not describe itself in the words "voluntary" or "mandatory", so the status should be read from the document and from the regulator's own statements.

In practice

Each dimension becomes a set of documents and controls. The first is a risk assessment with the limits on what the agent may do. The second is a named accountable person and the points where that person must approve or intervene. The third is the tool permissions, guardrails and logs. The fourth is the training and guidance given to the people who use the agent.

Example

[Composite] A government agency planned an agent to answer enquiries about licences. The risk assessment limited the agent to published guidance and barred it from changing records. The named accountable officer approved each new topic before the agent could answer it, and the logs were reviewed each month.

Common mistakes

• Treating the framework as a checklist completed once, rather than a cycle revisited after monitoring.
• Quoting an earlier version or a secondary summary instead of the version and date on the document.
• Describing the framework as a legal requirement without checking the regulator's own statement on its status.