MAS binds agents to bank rules

Singapore’s Monetary Authority said on August 5 that autonomous AI agents used by banks and other financial firms will fall within its forthcoming Guidelines on Artificial Intelligence Risk Management. The clarification came in a written parliamentary reply by Deputy Prime Minister and MAS Chairman Gan Kim Yong to questions from Member of Parliament Mariam Jaafar. MAS said the guidelines, first issued for consultation in November 2025, apply to all AI use cases by financial institutions, including agentic AI. The move gives Singapore a clearer supervisory position on agentic systems than many regulators that have so far focused on model governance or transparency duties. MAS has paired that position with a separate industry framework, published on July 3, called Safeguards for Agentic Finance at Runtime, or SAFR, which sets out how firms can govern agent actions before execution. (publicnow.com) ### What exactly did MAS say in Parliament? Mariam Jaafar asked for MAS’s assessment of the near-term risks from increasingly autonomous AI agents in financial services, whether MAS would move from the industry-led SAFR framework to mandatory supervisory requirements, and on what timeline. In reply, Gan said MAS would continue with a principles-based supervisory approach and that the AI risk guidelines will cover agentic AI together with other AI uses in financial institutions. (publicnow.com) The August 5 reply did not set out prescriptive technical rules for agents. Instead, MAS tied agentic systems to the same supervisory framework it has been building for AI risk management more broadly, with the regulator saying the guidelines will be finalized soon. ### What are banks being told to control? MAS and industry participants said on July 3 that AI agents in finance need “real-time safeguards” because they can act autonomously and at speeds beyond practical human intervention. (publicnow.com) The SAFR framework proposes governance checkpoints that verify and record an agent’s proposed actions before execution, with a focus on policy-bound execution, real-time validation, auditability and interoperability. Baker McKenzie said the SAFR paper introduces a governance checkpoint between each agent decision and execution so that no agentic action is carried out without being declared, authorized and assessed. The law firm also noted that SAFR itself is not regulatory guidance or a supervisory expectation, but a proposed runtime standard. (publicnow.com) ### How does this fit with Singapore’s broader AI policy? Singapore’s Infocomm Media Development Authority launched a Model AI Governance Framework for Agentic AI on January 22, saying agents can update customer databases or make payments and therefore require meaningful human control and oversight. That framework applies across sectors, while MAS’s work is specific to regulated financial institutions. (bakermckenzie.com) MAS has also been building implementation tools with industry. On March 20, the regulator said phase two of Project MindForge had produced an AI Risk Management Toolkit for the financial sector, developed with 24 banks, insurers, capital markets firms and other partners. (imda.gov.sg) ### What happens next for banks and fintechs? MAS said the AI risk guidelines launched for consultation on November 13, 2025 will be finalized soon, and Gan said the framework will remain principles-based as AI evolves. Separately, MAS said the Future of Finance Institute will support adoption of the SAFR framework through pilots and sandbox experimentation, while BuildFin.ai participants can help shape later iterations. (sgpc.gov.sg) For banks, insurers, capital-markets firms and fintechs licensed by MAS, the next formal milestone is publication of the final AI risk guidelines. For firms already testing autonomous workflows, the July 3 SAFR paper and the March 20 MindForge toolkit set out the controls MAS and industry participants are already discussing ahead of that release. (publicnow.com) (ebs.publicnow.com)