MAS proposes framework to govern AI agents

AI agents in finance now act faster than humans can directly intervene, MAS said.

The Monetary Authority of Singapore (MAS) is proposing an industry-developed framework for the use of artificial intelligence (AI) agents in financial services.

The Safeguards for Agentic Finance at Runtime (SAFR) paper emphasized the need for real-time safeguards that ensure that the AI agents’ behavior remain within predefined mandates, policies, and risk boundaries, MAS said in an announcement on 3 July.

Safeguards include policy-bound execution, real-time validation, auditability, and interoperability. The framework specifically sets out the direction for how the safeguards can be embedded into system operations for AI agents’ deployment, MAS said.

The need comes as AI agents in the financial services sector increasingly carry out tasks autonomously at a speed beyond practical human intervention, the central bank noted.

Amongst current AI agent applications in Singapore that have already applied the SAFR framework include in wealth management and advisory workflows, payments and treasury operations, and client engagement.

The AI agents review documents and generate assessments, execute routine transactions within predefined terms, or generate client insights with draft materials.

SAFR builds on MAS’ Project Mindforge’s AI Risk Management toolkit.