MAS updates safeguards for AI agent-driven financial transactions

The framework covers authorisation, transaction checks, and accountability.

The Monetary Authority of Singapore (MAS) has released updated safeguards for artificial intelligence (AI) agents that execute financial transactions.

The Safeguards for Agentic Finance at Runtime (SAFR) white paper, now in Version 1.1, includes contributions from an expanded group of financial institutions and frontier AI developers.

The framework proposes three measures — establishing agent identity and authorisation, checking proposed actions against users' instructions before execution, and maintaining accountability and auditability.

The move comes as AI agents could increasingly act on behalf of individuals and institutions to initiate and execute financial transactions, said MAS Managing Director Chia Der Jiun.

“Customer agents optimising and executing on behalf of individuals and companies could mean that money will move around faster for smaller differences in benefits and incentives,” Chia said at the INSEAD Digital Finance & Agentic AI Summit on 9 October.

"Financial institutions must be able to ensure that AI agents act within the scope of a user's instructions, within institutional policies and risk limits, and in compliance with applicable regulatory requirements," he added.

Moreover, using AI agents would require controls over authority, permissions, identity, accountability, and governance.

MAS has also released an open-source reference implementation of SAFR through the Future of Finance Institute to support industry experimentation, implementation, and further development.

The regulator invited research institutions, technology firms, and financial institutions to contribute to the framework's continued development.

"The technology is still developing rapidly, and many questions remain," Chia said.