MAS and Industry Partners Set Guardrails for AI Agents With New SAFR Framework

The Monetary Authority of Singapore has backed a new industry framework that sets tighter guardrails around AI agents operating inside financial systems. Developed alongside industry partners, the Safeguards for Agentic Finance at Runtime (SAFR) framework lays out how financial institutions can authorise AI agent actions, trigger human review, and maintain a full audit trail before any action is executed by an autonomous system.

SAFR responds to a growing regulatory need as financial institutions in Singapore and across the region begin deploying AI agents for tasks ranging from trade execution to customer onboarding. The framework introduces a runtime authorisation layer, meaning every action an AI agent attempts must pass through predefined checkpoints that can escalate to a human if the action falls outside approved bounds. This differs from pre-approval models, where agents are given blanket permission upfront.

The implications are significant for Singapore's financial sector, which is among the most AI-forward in Asia. With DBS, OCBC, and UOB already experimenting with AI agents across wealth management, fraud detection, and customer service, a standardised guardrail framework gives both incumbents and fintechs a clear compliance path. It also positions Singapore as a thought leader in agentic AI governance — an area where global regulators have yet to converge on consistent rules.

Why it matters for Singapore: As the city-state positions itself as a hub for agentic AI in financial services, SAFR gives MAS a template that could influence regional standards. Rather than letting individual institutions define their own risk boundaries in isolation, a coordinated framework reduces systemic risk while still allowing innovation to move quickly. For fintech companies building AI agents in Singapore, the message is clear: build safely from day one, or risk being locked out of the financial system.