Agentic AI governance framework updated with deployment case studies
IMDA updated the framework with industry feedback, multi-agent and third-party-agent practices, automation-bias guidance and real-world case studies.
Governance framework focused on agentic AI, including bounding autonomy, human accountability, lifecycle controls and transparency.
Organisations developing or deploying AI agents and agentic systems.
Voluntary governance framework.
IMDA updated the framework with industry feedback, multi-agent and third-party-agent practices, automation-bias guidance and real-world case studies.
Governance framework focused on agentic AI, including bounding autonomy, human accountability, lifecycle controls and transparency.
Set explicit boundaries for what agents can access, decide and execute, including tool and data permissions.
Actor: developers/deployers · Domain: agent controls
Evidence: Agent permission model, allowlists, budget/time limits and escalation controls.
Require human approval for high-impact, irreversible or materially consequential agent actions.
Actor: deployers · Domain: human oversight
Evidence: Approval matrix, intervention logs and exception records.
Apply controlled development, testing, deployment, monitoring and retirement practices to agents.
Actor: developers/deployers · Domain: vendor lifecycle
Evidence: Lifecycle gates, versioning, whitelisting and rollback capability.
Make agent roles, limitations and relevant automated actions understandable to users and accountable operators.
Actor: developers/deployers · Domain: transparency
Evidence: User disclosures, action traces and operator guidance.
Cybatar's mapping is designed for AI governance and assurance work. Legal interpretation remains anchored to the current primary text and competent authority guidance.
Open IMDA Singapore source