Not sure where to start? Answer 3 questions and get a tailored compliance action plan.
What applies to me? →Model AI Governance Framework for Agentic AI, Version 1.5
Issued by
Infocomm Media Development Authority (IMDA)
Singapore's IMDA released Version 1.5 of its Model AI Governance Framework for Agentic AI, extending voluntary guidance for organizations developing or deploying autonomous AI agents in enterprise environments. The update addresses how enterprises should scope agent permissions, maintain human oversight, and manage agent lifecycles from deployment through decommissioning. It applies to any organization operating AI systems capable of taking goal-directed actions with limited human intervention.
Applies To
Overview
Version 1.5 builds on the foundational agentic AI governance framework published by IMDA, refining guidance in response to observed enterprise deployment patterns and emerging risks from multi-agent architectures. The update introduces more granular recommendations for defining and enforcing agent permission boundaries, specifying what actions an agent may initiate autonomously versus those requiring human authorization. It also strengthens lifecycle governance expectations, covering agent onboarding, monitoring, versioning, and retirement. Enforcement of this framework remains voluntary, consistent with Singapore's broader model governance approach, though adoption is strongly encouraged and may inform future regulatory baselines. Organizations operating in regulated sectors such as financial services or healthcare may face additional obligations from sectoral regulators that reference or align with this framework.
Key Requirements
- •Define and document explicit permission scopes for each deployed agent, specifying authorized action categories and hard boundaries.
- •Establish human oversight checkpoints for high-consequence or irreversible agent actions, with escalation procedures documented.
- •Maintain a lifecycle log covering agent deployment, configuration changes, and decommissioning events.
- •Conduct periodic reviews of agent behavior against intended objectives, with findings recorded and acted upon.
- •Ensure multi-agent systems include inter-agent accountability mechanisms that attribute actions to identifiable components.
- •Assess and document residual risks before deploying agents in customer-facing or critical operational contexts.
What Your Organization Must Do
- →Audit all currently deployed AI agents and map their permission scopes against the updated framework criteria.
- →Update internal agent deployment policies to incorporate the new lifecycle governance checkpoints specified in Version 1.5.
- →Establish or revise escalation workflows so human operators can intervene when agents approach defined action boundaries.
- →Review vendor and partner contracts involving third-party agentic AI tools to confirm lifecycle accountability provisions are addressed.
- →Assign ownership for ongoing agent monitoring to a named role or team, with a documented review cadence.
- →Brief senior risk and legal stakeholders on how sectoral regulators in Singapore may reference this framework when assessing agentic AI deployments.
Playbook Guidance
Step-by-step implementation guidance for compliance teams.
Frequently Asked Questions
- Is Singapore's Model AI Governance Framework for Agentic AI Version 1.5 legally binding?
- No, compliance with MAIGF-AA1.5 is voluntary. IMDA positions it as guidance rather than regulation, though adoption may inform future regulatory baselines and sectoral regulators in financial services or healthcare may reference it when assessing agentic AI deployments.
- What counts as an 'agentic AI system' under IMDA's Version 1.5 framework?
- The framework applies to any AI system capable of taking goal-directed actions with limited human intervention. This covers both single-agent deployments and multi-agent architectures operating autonomously in enterprise environments.
- What specific documentation does Version 1.5 require organizations to maintain for deployed AI agents?
- Organizations should maintain a lifecycle log recording deployment events, configuration changes, and decommissioning. They must also document agent permission scopes, human oversight escalation procedures, and periodic behavioral review findings.
- How does MAIGF-AA1.5 handle multi-agent systems differently from single-agent deployments?
- Version 1.5 introduces inter-agent accountability requirements specific to multi-agent architectures, requiring organizations to ensure actions can be attributed to identifiable system components. This addresses emerging risks from coordinated autonomous agent behavior that single-agent guidance did not fully cover.
- Do Singapore financial services firms face additional obligations beyond MAIGF-AA1.5 for agentic AI?
- Yes. Organizations in regulated sectors such as financial services or healthcare may face obligations from their sectoral regulators that align with or reference this framework, meaning MAIGF-AA1.5 effectively sets a practical compliance floor for those industries.
- When is the effective date for MAIGF-AA1.5 and what should compliance teams prioritize before then?
- The framework carries an effective date of May 20, 2026. Compliance teams should prioritize auditing existing agent deployments, mapping permission scopes, updating lifecycle governance policies, and reviewing third-party vendor contracts involving agentic AI tools before that date.