The Monetary Authority of Singapore (MAS) has published an industry white paper proposing a runtime governance framework for AI agents operating in financial services, marking a shift from static model oversight towards controls that operate at the point an autonomous system acts.
The paper, Safeguards for Agentic Finance at Runtime (SAFR), was developed with financial institutions and FinTech firms under MAS’ BuildFin.ai initiative, which supports the responsible development and deployment of artificial intelligence in the financial sector. MAS says the framework is designed to enable AI agents to carry out financial tasks “safely, securely and reliably”.
The operating issue is that agentic AI systems can initiate or complete tasks at a speed and scale that makes manual intervention impractical. SAFR responds by defining governance checkpoints that verify and record an AI agent’s proposed actions before execution, keeping activity within the mandates, policies and risk limits set by the financial institution.
For compliance, risk and technology teams, the framework points to a more operational form of AI assurance. Controls such as policy-bound execution, real-time validation, auditability and interoperability are embedded into workflows, rather than applied only through pre-deployment review or post-event monitoring. The practical effect is to shift governance closer to the execution layer, where an AI agent requests authority to act.
Industry participants have tested the approach across payments and treasury operations, wealth management and compliance review, and client engagement. Reported examples include agents handling routine payments and treasury transactions within predefined limits, reviewing documents and generating structured compliance assessments, and drafting client materials within approved content boundaries.
SAFR builds on MAS’ Project MindForge AI Risk Management Toolkit by focusing on how safeguards can be operationalised at the point of action for AI agents. The framework also extends MAS’ BuildFin.ai work by moving responsible AI deployment into live system operations, where agent actions can be authorised, validated and recorded before execution.
SAFR is significant because it treats agentic finance as an execution-risk problem as much as a model-risk problem. Traditional AI governance frameworks have focused on inputs, outputs, explainability, testing and accountability. Agentic systems add a further control question: whether the system should be allowed to take a specific action, in a specific context, at a specific point in time.
That has direct implications for RegTech architecture. Firms deploying AI agents in regulated workflows will need evidence that actions were authorised, exceptions were escalated, and decision records can be reconstructed for audit, supervision and incident review. MAS has invited further industry participation in future SAFR iterations, while the Future of Finance Institute will support adoption through industry pilots and sandbox experimentation.
Subscribe to our newsletter
