Singapore's MAS Proposes Safeguards for AI Agents in Finance

The Monetary Authority of Singapore has proposed a framework to govern the use of autonomous artificial-intelligence agents in financial services, an early move to set guardrails as banks, insurers and asset managers experiment with software that can take actions on their own. The proposed approach, described as safeguards for agentic finance at runtime, seeks to define how an AI agent's actions are authorized, the circumstances under which human oversight must be activated, and what information should be recorded so that decisions can be traced and audited after the fact. Agentic AI goes beyond tools that simply generate text or recommendations, instead carrying out multi-step tasks such as executing transactions, adjusting portfolios or handling customer requests with limited human intervention, which raises fresh questions about accountability, control and consumer protection. By focusing on runtime controls, the framework aims to ensure that firms retain meaningful oversight of what their AI systems actually do in live operations, not just how they are designed and tested. The proposal reflects Singapore's ambition to stay at the forefront of financial innovation while managing emerging risks, and adds to a growing body of global work on how to supervise increasingly capable AI in finance.