Skip to content

Singapore's Central Bank Wants Banks to Check Who an AI Agent Is Before It Pays

MAS proposed a know-your-agent protocol and an Agentic Money Sandbox to test stablecoin payments by AI agents, with identity and authority checks at the bank before execution.

Category
Industry
Published
Author
A2AWire
Reading time
2 min read

A regulator has finally put agent payments on the exam table. At the Singapore FinTech Festival on Nov. 19, Monetary Authority of Singapore (MAS) managing director Chia Der Jiun outlined a plan for banks to verify who an AI agent represents and what its customer authorized before any money moves. A day earlier, MAS and the Future of Finance Institute announced a planned Agentic Money Sandbox to test those safeguards, with stablecoins and tokenized deposits among the payment instruments.

Highlights#

Under the proposed know-your-agent protocol, a financial institution would receive details on the agent's identity, the customer behind it, its delegated authority, and any assurances about its origin and reliability. Banks could then scope the agent's access and route it through a dedicated secure channel.

The payments trial would place verification at the bank or payment provider: the agent submits an instruction, the institution checks identity and authority and applies risk controls, then executes using tokenized deposits or stablecoins.

A separate corporate treasury trial would let an agent monitor balances and upcoming payments, decide when and where funds move, and carry out permitted transfers. The company sets the mandate; the bank sets its own terms for accepting each instruction.

Launch partners named in the joint release include DBS, OCBC, Standard Chartered, and UOB. No sandbox start date or participant list was given. Chia also left open the hardest question: who is responsible when an agent makes a mistake or exceeds its mandate.

Why it matters#

This is the first time a major central bank has sketched a verification layer for machine-initiated payments, and it lands directly on the problems A2AWire works on: proving agent identity, encoding spending mandates, and settling in stablecoins with controls at the point of execution rather than after the fact.

Sources#

Related news

On A2AWire