Singapore's MAS Proposes Runtime Safeguards for AI Agents in Finance
A new framework developed with eight global financial firms aims to verify AI agent permissions before payments or trades are executed.
The Monetary Authority of Singapore (MAS) has released a white paper titled "Safeguards for Agentic Finance at Runtime" (SAFR) to establish a technical governance layer for AI agents. The framework proposes that financial institutions verify an AI agent's identity, permissions, and risk limits in real-time before any payment or trade is executed.
Developed in collaboration with eight major firms—Ant International, Circle, HSBC, JPMorganChase, Manulife, Mastercard, OCBC, and Visa—the SAFR architecture is built on four core components: Agent Identity, a Controls Repository, a Disposition Engine, and an Audit Log. Under this system, AI-driven actions result in one of four specific dispositions: Auto-Execute, Escalate, Observe, or Deny. While the proposal provides a concrete operational blueprint, MAS has clarified that SAFR is currently an industry reference and does not constitute binding regulatory guidance.
The Shift to Runtime Governance
This initiative arrives as the volume of AI-generated financial decisions begins to outpace the capacity for manual compliance reviews. Traditionally, financial oversight has relied on reactive audits to catch errors after they occur. However, as AI agents gain the autonomy to move capital and execute complex trades, the risk of catastrophic financial errors or fraud has increased.
Singapore's approach contrasts with other global regulatory trends. While the European Union has focused on enhanced governance and the UK's Financial Conduct Authority (FCA) is moving toward a model of "stewardship," Singapore is attempting to build a specific technical infrastructure. Nikhil Rathi, CEO of the FCA, has noted that legislation will never keep up with the pace of AI, suggesting that regulators must intervene on judgment rather than waiting for detailed rules in every instance.
Industry Implications
By requiring a digital "permission slip" before an action is finalized, SAFR shifts the industry toward proactive prevention. This runtime governance model aims to stop unauthorized or high-risk transactions before they hit the ledger, reducing the reliance on post-event recovery.
Private sector players are already moving toward similar controls. In May 2026, Fiserv launched "agentOS," a dedicated environment for banks to deploy AI agents equipped with built-in audit trails and kill switches. The convergence of regulatory white papers and commercial products suggests a broader industry consensus that autonomous finance cannot function without hard-coded guardrails.
Future Outlook
As the SAFR framework moves from a white paper to potential implementation, the industry will be watching to see if other jurisdictions adopt similar technical standards. The primary challenge remains the balance between the efficiency of autonomous agents and the necessity of human oversight. Whether these safeguards become a global standard for "agentic finance" will depend on how effectively banks can integrate the Disposition Engine into existing legacy payment systems without creating prohibitive latency.