The Infocomm Media Development Authority of Singapore (IMDA) and the Monetary Authority of Singapore (MAS) have published the Model AI Governance Framework for Agentic AI and the Guidelines on Artificial Intelligence (AI) Risk Management, respectively. Both publications promote the responsible adoption of AI by providing guidance on governance, oversight, and risk management. The former specifically addresses agentic AI systems, while the latter sets out supervisory expectations for financial institutions.


Singapore’s Agentic AI Governance and AI Risk Management

Watch video

Executive summary

The IMDA Model AI Governance Framework for Agentic AI, which is voluntary and cross-sectoral in scope, establishes four dimensions for the responsible development and deployment of AI agents: prior risk assessment and definition of boundaries, effective human accountability, implementation of technical controls and processes, and empowerment of end users to ensure responsible use.

Meanwhile, the MAS Guidelines on AI Risk Management set out supervisory expectations for financial institutions regarding AI governance and oversight, the identification and inventory of AI systems, the assessment of risk materiality, controls throughout the AI lifecycle, and capabilities and resources, including technological infrastructure.

The MAS Guidelines will be implemented in phases. Financial institutions must meet the expectations relating to AI governance and oversight, as well as AI risk management systems, policies and procedures, no later than 7 October 2027. Expectations concerning AI lifecycle controls and capabilities and resources must be met no later than 7 October 2028.

Main content

The Framework establishes four key dimensions for the responsible development and deployment of agentic AI:

  • Assess and bound risks upfront. Identify risks arising from agent autonomy and establish boundaries for agents' actions, permissions and access to data and systems.
  • Make humans meaningfully accountable. Define clear roles and responsibilities and incorporate human oversight mechanisms for critical decisions and actions.
  • Implement technical controls and processes. Implement safeguards during development, conduct pre-deployment testing and maintain continuous monitoring of agents.
  • Enable end-user responsibility. Provide information and training on agents' capabilities, limitations and risks to promote responsible use.

The MAS Guidelines establish three main areas of supervisory expectations for financial institutions:

  • Governance, oversight, and capabilities. Define the responsibilities of boards of directors and senior management and ensure appropriate oversight of AI risks and maintain adequate capabilities and resources, including technological infrastructure.
  • Key systems, policies and procedures for AI risk management. Establish mechanisms for AI identification and inventory management, and assess risk materiality to apply proportionate controls.
  • Lifecycle controls. Implement measures covering data management, transparency, fairness, human oversight, evaluation and testing, cybersecurity, and third-party risk management.

Access the technical note on the Singapore’s Agentic AI Governance and AI Risk Management.