MAS Proposes New SAFR Framework to Govern AI Agents in Finance

The Monetary Authority of Singapore (MAS) has proposed a new industry-developed framework for governing artificial intelligence agents in financial services, responding to a landscape where AI systems now execute decisions faster than humans can meaningfully intervene. The framework, called Safeguards for Agentic Finance at Runtime (SAFR), lays out how real-time guardrails can be embedded directly into system operations rather than applied as after-the-fact compliance checks.

MAS identified four core safeguard categories: policy-bound execution that keeps AI agents within predefined mandates and risk boundaries; real-time validation to catch out-of-bounds behaviour before it settles; auditability so every agent action can be traced and reviewed; and interoperability standards so different agent systems can operate within the same governance envelope. The framework builds directly on MASS earlier Project Mindforge, which produced an AI Risk Management toolkit for financial institutions.

Several Singapore financial institutions have already begun applying SAFR principles to live workflows. Current use cases span wealth management and advisory processes where AI agents review documents and generate preliminary assessments, payments and treasury operations where agents execute routine transactions within predefined parameters, and client engagement workflows where agents generate personalised insights and draft materials. MAS noted that these early adopters demonstrate the frameworks practical viability across different banking functions.

The central banks move comes as AI agents proliferate across Singapores financial sector, from automated trading systems to robo-advisory platforms and fraud detection chains. Without governance guardrails built into the runtime layer, MAS warned, the speed advantage of AI agents becomes a systemic risk rather than an efficiency gain. The SAFR framework is designed to be technology-neutral and adaptable as agent capabilities evolve, avoiding prescriptive rules that would quickly become obsolete.

Why it matters for Singapore: MASS SAFR proposal positions Singapore as one of the first major financial centres to tackle the governance of autonomous AI agents at the infrastructure level rather than through surface-level policy. With Singapore already a global fintech hub processing trillions of dollars in transactions, getting agentic AI governance right is not just a regulatory exercise it is a competitive advantage. Financial institutions operating out of Singapore will be the first to benefit from clear, workable guardrails, while those elsewhere may face regulatory uncertainty as they deploy similar technologies.