Singapore's MAS Proposes Guardrails for AI Agents in Finance

The Monetary Authority of Singapore has proposed a governance framework for the use of autonomous artificial intelligence agents in financial services, reflecting the growing role such systems could play in banking, insurance and investing. The framework, referred to as Safeguards for Agentic Finance at Runtime, or SAFR, sets out how the actions of AI agents should be authorised, how and when human oversight is activated, and what information must be recorded at the point of every decision. Unlike conventional AI models that primarily generate analysis or recommendations, agentic systems can take actions on their own, such as executing tasks or transactions, which raises new questions about accountability, control and auditability. The proposal aims to ensure that as financial institutions adopt these tools, there are clear guardrails around authorisation, escalation to humans and record-keeping, so that risks can be managed and decisions traced. It also discusses practical considerations for implementation. The move is consistent with Singapore's broader, principles-based approach to responsible AI in finance, positioning the city-state to encourage innovation while maintaining trust, safety and effective supervision as agent-driven technology matures.