This page may contain stale information. Last updated: 2026-07-05

Definition

SAFR (Safeguards for Agentic Finance at Runtime) is an industry-developed framework published by monetary-authority-of-singapore and financial partners (July 2026) for real-time governance of AI agents in financial services.

Architecture

SAFR sits between the agent and systems it acts on — a Governance Envelope evaluating proposed actions before execution:

ComponentFunction
Agent identityWho/what is proposing the action
Controls repositoryPolicies, mandates, risk thresholds
Disposition engineApprove, reject, escalate to human, or flag for monitoring
Audit logTamper-evident record of governance decisions

Implementation Models

  • Native integration: Agent produces governance record before each proposed action
  • Gateway model: Intercepts outbound API calls from existing agents

Distinction from Static Governance

Shifts from training-time / policy-document governance to action-time verification — the missing runtime layer as agentic finance scales beyond demos.

Sources