0x2a9d73d92a9d…2a9d73dc

ConfirmedRegulation & Policy497 vB79 sat/vB2 min decode

Singapore Plans Agentic Money Sandbox for AI-Run Payments

MAS and FFI plan an Agentic Money Sandbox letting banks test AI agents that initiate digital-currency payments, with Purpose Bound Money setting conditions, caps and accountability.

Singapore Plans Sandbox to Test AI-Run Digital-Currency Payments
WitnessSingapore Plans Sandbox to Test AI-Run Digital-Currency PaymentsJESHOOTS-com / pixabay

Outputs

  1. MAS and FFI are planning an Agentic Money Sandbox for banks to test AI agents initiating digital-currency payments

  2. Five banks are testing AI to identify scam funds moving between institutions

  3. A design study for an industrywide scam detection system begins in early 2027

  4. Purpose Bound Money would set conditions, permitted uses and spending caps before funds move

  5. The sandbox covers scam detection, small-business financing and automated transactions

The Monetary Authority of Singapore (MAS) and the Foundation for Applied Innovation (FFI) are planning a regulatory sandbox in which banks can test artificial-intelligence agents that initiate digital-currency payments on their own, while the participating businesses and financial institutions remain legally responsible for every transaction. The sandbox, provisionally named the Agentic Money Sandbox, is still in the planning and testing stage, according to the proposal's outline covering identity checks, permission controls, transaction tracking and transaction responsibility.

The program targets three areas of financial AI: scam detection, small-business financing and automated transactions. In the sandbox, banks would test agents that manage funds and execute transactions between financial institutions — a step beyond today's AI deployments, which largely stop at analytics and fraud flagging.

What does the sandbox change for practitioners?

The operational core of the proposal is Purpose Bound Money, a programmable-money construct that defines conditions, permitted uses and spending caps before any funds move. Under the design, an AI agent would verify whether contractual duties had been met and then release payment under preset rules — with no manual handling of each individual transaction.

For treasury and payments teams, that shifts the compliance question from "who clicked approve" to "who coded the conditions." The proposal keeps accountability on the regulated side: the institutions, not the agents, answer for every transfer. Proposed test parameters include identity verification, permission structures, transaction tracking and the allocation of responsibility for completed payments.

A parallel track against scam flows

Separately from the sandbox, five banks are already testing AI systems to identify scam funds moving between institutions — an interbank detection effort rather than a single-bank control. The detection problem is material: stolen funds typically hop across multiple banks within minutes, and per-institution monitoring sees only one leg of the movement.

Two further commitments frame the timeline. Singapore plans to provide synthetic data to eligible developers, addressing the scarcity of labeled scam-transaction data that limits model training. And an industrywide scam detection system will begin a design study in early 2027 — the first fixed milestone practitioners can plan against.

The financing pilot

The third workstream, small-business financing, will assess whether AI can use bank statements and invoices to underwrite businesses with limited credit histories or collateral. For such firms, document-based assessment could substitute for the formal credit files that traditional underwriting requires; for banks, it opens a segment historically judged too expensive to assess manually.

The plan stops short of production authorization. Banks entering the sandbox test under controlled conditions, and the stated scope — identity checks, permission controls, tracking, responsibility — reads as a mapping exercise for where liability sits when an autonomous agent moves funds without a human sign-off at each step.

The design study for the industrywide scam detection system in early 2027 marks the next observable checkpoint; until then, the five-bank detection trial and the sandbox's rule-setting phase define the practical frontier for AI-initiated payments in Singapore.

via tokenpost.com (Source)

More from Nathan Brooks

Nathan Brooks

Show full bio

Market editor covering money, funding and market shifts around crypto and web3 industry at Mempool Brief.

1,512 articles