Singapore Model AI Governance Framework for Agentic AI
Practical guidance for organizations deploying AI agents that can take actions on people's behalf.
Published by Infocomm Media Development Authority
How to use it
The framework applies governance practices to agents that can operate tools or carry out a series of steps. It emphasizes limiting agent permissions, retaining human accountability, and using controls throughout deployment.
For an agent that works with company email, separate the ability to read messages from the ability to send them. Set an approval requirement for outgoing mail, make actions visible to the operator, and test how the agent stops when a request falls outside its allowed scope.
- Access
- Free publication from IMDA