Summary
The Monetary Authority of Singapore (MAS), together with Mastercard, Visa, Circle, OCBC, Bank of Singapore, Ant International, and Manulife, launched the SAFR (Safeguards for Agentic Finance at Runtime) framework — a first-of-its-kind industry reference model for governing AI agent actions in financial systems. SAFR sits between an AI agent and the systems it acts on, checking identity, authority, controls, and risk thresholds before any action executes.
Key Facts
- SAFR: Safeguards for Agentic Finance at Runtime — developed under MAS' BuildFin.ai initiative
- Four components: agent identity, controls repository, disposition engine, audit log (tamper-evident)
- Disposition engine determines: approve, reject, human review, or flag for monitoring
- Industry pilots: Mastercard, Visa, Circle, OCBC, Bank of Singapore, Ant International, Manulife
- Use cases: agent-assisted payments, treasury operations, wealth management, client engagement
- Not regulatory guidance — presented as adaptable industry reference model
- Two implementation paths: native integration (agent produces governance record) or gateway model (intercepts API calls)
- Builds on MAS' Project MindForge AI Risk Management toolkit
- Future of Finance Institute will support adoption through industry pilots and sandbox experimentation
Why It Matters
SAFR is the first regulatory-backed framework specifically designed for AI agents that can take action (not just recommend). As AI agents increasingly initiate payments, submit trades, approve credit, and settle claims, existing governance processes built for human decision-making are inadequate. Singapore is positioning itself as the global standard-setter for agentic AI governance in finance.