Summary
- Launched by Singapore's Minister for Digital Development and Information, Josephine Teo, at the World Economic Forum on 22 Jan 2026. Developed by IMDA, building on the Model AI Governance Framework introduced in 2020.
- Aimed at organisations deploying agentic AI, whether they build agents in-house or buy third-party agent products.
- Why it exists: agents can act on their own (update a database, make a payment), which adds risks such as unauthorised or wrong actions and automation bias (over-trusting a system that usually works).
- Four dimensions (from IMDA's release and factsheet):
- Assess and bound risks upfront: pick suitable use cases using agent-specific factors like data access and autonomy, then limit autonomy, tools and data by design.
- Make humans meaningfully accountable: assign responsibility across all parties, require human approval at significant checkpoints, and audit whether those approvals actually work.
- Implement technical controls and processes: controls for planning, tools and protocols; baseline safety and security testing before deployment; gradual rollout with continuous monitoring.
- Enable end-user responsibility: tell users when and how agents are used, and train them to oversee agents while keeping their own core skills.
- IMDA says it is also developing guidelines for testing agentic AI applications.
Who it applies to
- Any organisation in Singapore (or elsewhere) that chooses to follow it. No legal force.
- Not extraterritorial.
Key dates
- Earlier Model AI Governance Framework (MGF for AI) introduced, which this builds on
- Minister Josephine Teo launches the MGF for Agentic AI at the World Economic Forum
Full date record
| Date | Event |
|---|---|
| 2020 | Earlier MGF for AI introduced |
| 2026-01-22 | MGF for Agentic AI launched at WEF |
Penalties and enforcement
- None. Voluntary, with IMDA inviting feedback and case studies.
What it means for an Indian company
- The most practical public checklist for AI agents today, and highly relevant to Indian SaaS and IT services firms building agents for global clients.
- Map it to controls: risk bounding to SC-AI-004 AI risk classification and use-case tiering and SC-AI-030 Access control for AI systems and agents (least privilege for agents' tools and data); human checkpoints to SC-AI-026 Human oversight and override; testing and phased rollout to SC-AI-015 Pre-deployment testing and validation and SC-AI-020 Performance monitoring and drift detection; agent attacks to SC-AI-029 AI-specific security controls (prompt injection, model abuse); user notice to SC-AI-022 Disclose AI interaction to users; staff training to SC-AI-008 AI literacy and staff training.
- Indian banks and fintechs using agents should read it next to RBI FREE-AI, which also stresses human oversight.
Obligations
Full obligation breakdown coming
We break the most requested policies down into numbered obligations mapped to controls. Leave your details and we’ll tell you when Singapore MGF for Agentic AI is done.
Glossary
- Agentic AI / AI agent: an AI system that can plan and take actions to finish a task for a user, not just produce text.
- Automation bias: the habit of trusting an automated system too much because it worked before.
- IMDA: Infocomm Media Development Authority, Singapore's digital regulator and developer.
- MGF: Model AI Governance Framework, Singapore's series of voluntary AI governance guides.
- Least privilege: giving a user or agent only the access it needs for the task.
Key takeaways
- Voluntary Singapore framework for deploying AI agents, launched 22 Jan 2026.
- Four dimensions: bound risks, human accountability, technical controls, user responsibility.
- "Active (Voluntary)" matches.
- A strong template for agent controls in any SecComply client using agents.
Official sources
- IMDA press release, 22 Jan 2026, accessed 2026-10-19
- IMDA factsheet: MGF for Agentic AI (PDF), accessed 2026-10-19
- IMDA AI page, section on the MGF for Agentic AI (framework document), linked from the press release; not opened
This is general information, not legal advice. Last verified 19 Oct 2026 against the official source. Laws change; confirm the current position with qualified counsel before acting.