Singapore issues agentic AI governance guide to curb autonomy risks

The playbook calls for tool and data access limits, approval gates, lifecycle controls, plus transparency and staff training.

Singapore has launched a new Model AI Governance Framework for Agentic AI, becoming the first jurisdiction to issue a comprehensive enterprise guide for the responsible deployment of autonomous AI agents, according to the Infocomm Media Development Authority (IMDA).

IMDA said agentic AI systems offer productivity gains by automating tasks such as customer service and enterprise workflows, but also introduce new risks because of their autonomy and access to sensitive data.

These risks include unauthorised actions, errors, and challenges in maintaining effective human accountability, particularly where users over-rely on automated systems

The framework sets out guidance across four areas, including limiting agent autonomy and access to tools and data, defining checkpoints requiring human approval, implementing technical controls throughout the AI lifecycle, and strengthening user responsibility through transparency and training.

IMDA said the measures aim to ensure meaningful human oversight and accountability in agentic AI deployment.

IMDA said it developed the framework with input from government agencies and private sector organisations, and described it as a living document that will evolve with feedback and real-world use cases.

The authority is also developing additional testing guidelines for agentic AI applications, building on its existing safety and reliability tools for large language models.