Singapore has unveiled a new governance framework aimed at managing the risks associated with autonomous or agent-based artificial intelligence systems. The framework, released on 22 January 2026 by the Infocomm Media Development Authority (IMDA), focuses on AI systems that are capable of planning, reasoning and executing tasks with limited human input. These so-called AI agents are increasingly used in areas such as workflow automation, decision support and digital services.
Unlike conventional AI models that operate strictly in response to prompts, agentic AI systems can initiate actions independently. This creates new risk considerations, including unintended system access, misuse of sensitive data, and decisions that may be difficult to reverse once executed. Singapore’s new framework addresses these concerns by setting out governance principles that organisations are expected to adopt when deploying such technologies.
A core element of the framework is the requirement for clear boundaries on what AI agents are allowed to do. Organisations are encouraged to define access controls, restrict connections to critical systems, and establish approval checkpoints for actions that could have significant operational or legal consequences. The framework also reinforces that accountability remains with human decision makers and organisations, even when tasks are delegated to autonomous systems.
The guidance further highlights the importance of monitoring, auditability and fallback mechanisms so that AI agent behaviour can be reviewed and corrected when necessary. By issuing this framework at an early stage of agentic AI adoption, Singapore aims to provide practical guardrails that allow innovation to continue while reducing the likelihood of harmful or uncontrolled outcomes.
The initiative reflects Singapore’s broader strategy of shaping trusted AI development through forward-looking governance rather than reactive regulation.
Official source from IMDA – MODEL AI GOVERNANCE FRAMEWORK FOR AGENTIC AI: