Singapore plans agent identity protocol and digital-money settlement
Singapore’s MAS plans a know-your-agent protocol, an Agentic Money Sandbox and an SG Digital Ledger after settling $100 million in MAS Bills against a Singapore dollar central bank digital currency.

Why it matters
- MAS announced its know-your-agent protocol and SG Digital Ledger plans on November 19, 2026.
- MAS and the Future of Finance Institute announced the Agentic Money Sandbox on November 18, 2026.
- $100 million in MAS Bills was issued and settled against a Singapore dollar central bank digital currency in November 2026.
- SAFR, published July 3, 2026, set out four lessons for authorising and overseeing agents.
- BLOOM brings together 24 industry members and policymakers from the region.
MAS said $100 million in MAS Bills was issued and settled against a Singapore dollar central bank digital currency in November 2026.
On November 19, 2026, the Monetary Authority of Singapore (MAS) said it would help develop a know-your-agent protocol, a mechanism for verifying an AI agent’s identity and authority.
The authority also plans to reposition its SGD Testnet as the SG Digital Ledger, an operational settlement utility for institutions, tokenised assets, several forms of digital money and separate ledgers.
Managing Director Chia Der Jiun announced both plans at the Singapore FinTech Festival, where some payment service providers and financial institutions are already building agent interfaces.
MAS said many institutions have yet to establish terms or channels for agent access. Its response combines access rules with the settlement infrastructure required for agentic transactions, or transactions initiated and carried out by AI agents.
Takeaway: MAS’s plans target infrastructure gaps before autonomous agents receive broad access to customer funds.
Who controls an agent when it moves money?
Chia warned that granting an agent a customer’s login credentials creates insufficient safety assurance. “Some users are giving agents their financial-account login credentials,” he said, advising against the practice.
MAS said institutions lack certainty over three linked questions: whether the customer authorised the agent, whether it stayed within its mandate and instructions, and who bears responsibility if it exceeded them.
Under MAS’s proposal, the protocol would support trusted information exchange between customer agents and financial institutions. Each exchange would establish:
- The agent’s identity and the person or organisation it represents.
- The authority it received and the mandate, meaning its permitted actions and limits.
- The assurance available about its provenance, meaning where it came from, and its reliability.
Under the proposal, financial institutions could use that record to set each agent’s terms and permitted scope. They could also establish a dedicated secure channel for agent access.
MAS stressed that the protocol remains under development and will require industry collaboration on common approaches. The November 19 announcement therefore described work toward a shared access mechanism, not a finished standard ready for deployment.
MAS said the plan builds on Safeguards for Agentic Finance at Runtime (SAFR), published July 3, 2026. Financial-industry participants developed SAFR jointly under MAS’s BuildFin.ai initiative.
SAFR addresses how agent actions are authorised, how human oversight is activated and what gets recorded at each decision. Its first phase produced four lessons:
- Agents need bounded autonomy.
- Identity and authority must be established.
- Mandates must specify permissible actions, limits and conditions.
- Controls should strengthen as an agent’s discretion and potential impact increase.
MAS said those principles now inform the proposed exchange of identity, authority and mandate information.
Takeaway: The protocol’s core test is whether banks can prove authority before an agent acts, not merely after it acts.
What can banks test in the Agentic Money Sandbox?
MAS and the Future of Finance Institute separately announced the Agentic Money Sandbox on November 18, 2026. The controlled environment lets financial institutions develop and test agents that transact with digital money under defined authority, limits and safeguards before live deployment.
The sandbox combines AI agents with programmable money, meaning money carrying rules for automated execution. MAS said its initial use cases are agent-initiated payments and corporate treasury management.
MAS said a treasury agent could monitor liquidity positions and due payments. It could decide when and where funds should move, then use tokenised deposits, or digital representations of bank deposits, for permitted actions.
Under MAS’s sandbox design, the corporate would set the mandate. Participating banks would establish controls and the terms under which they accept each instruction.
For payments, MAS said banks and payment service providers would verify an agent’s identity and authority before execution. They would also apply risk controls to transactions using tokenised deposits or stablecoins, which MAS treats as a regulated form of digital money.
MAS did not report completed live deployments, transaction volumes or performance results from the sandbox. The announcement described a test environment, not evidence that agents can yet move funds safely at scale.
Takeaway: The sandbox can expose control failures before deployment, although MAS has published no measured results from it.
How does the SG Digital Ledger change settlement?
MAS’s SG Digital Ledger carries the same scale-up agenda into settlement, where the existing testnet has already supported interbank settlement and MAS Bill issuance. MAS reported $100 million in MAS Bills settled against a Singapore dollar central bank digital currency in November 2026.
The ledger’s intended scope, according to MAS, spans institutions, tokenised assets, different forms of digital money and separate ledgers. The announcement did not provide a common rulebook for all those connections or a timetable for operational launch.
Chia made adoption explicitly conditional: “Market demand would determine how widely the utility is adopted,” he said. Chia’s statement places the burden on actual market use rather than an automatic switch from the testnet.
MAS gave no adoption forecast alongside the announcement. That leaves open basic commercial questions about which institutions would connect, what assets would settle and how quickly the utility could support wider tokenised markets.
Takeaway: The ledger has a completed $100 million settlement test, but its wider use still depends on market demand.
What infrastructure guidance accompanies the plans?
MAS released two publications alongside Chia’s keynote. The Guardian Tokenised Collateral Framework, developed by industry workgroup members, sets operational requirements covering which collateral, meaning assets pledged to support financial obligations, is eligible, plus custody, valuation, return and enforcement.
MAS said those arrangements matter because collateral supports derivatives, secured funding and liquidity management. The framework covers how assets qualify for use, where institutions keep them, how they are valued, how returns work and what happens when enforcement is required.
The BLOOM Field Guide to Extending Settlement Capabilities examines functions, structure, connectivity and controls across end-to-end digital money transactions. It covers different institutional, technical and jurisdictional settings rather than one market’s implementation.
BLOOM brings together 24 industry members and policymakers from the region. MAS sees tokenised bank deposits, regulated stablecoins and wholesale central bank digital currency as possible forms of settlement money.
MAS presented the publications as operational guidance rather than adoption evidence. It did not provide an implementation timetable or say that the proposed rules had already been adopted across the market.
Takeaway: The two documents fill specific infrastructure gaps, but guidance alone cannot establish market adoption.
What gains and risks did Chia identify?
Chia said agents could deliver four potential economic benefits:
- Lower search and information costs.
- Continuous comparison or negotiation.
- Automated procurement.
- More dynamic pricing.
Chia said these are potential consequences of broader adoption, not measured outcomes of the protocol, sandbox or ledger. The announcement supplied no figures for the savings, pricing changes or productivity gains.
He also cited emerging studies in simulated environments suggesting that agents could collude without instructions to do so. More capable agents could secure better prices and terms, although MAS named neither the studies nor their methods in the announcement.
Chia said those outcomes could raise competition concerns and the risk of unequal access to capable models. Faster agent-directed movement of money and asset allocation could also prompt financial institutions to consider transaction limits.
Regulators may also need to assess the financial-stability effects of that speed, according to Chia. His stated boundary was that consumers and businesses retain meaningful control over agents’ objectives, mandates and actions.
Chia connected the protocol, sandbox and ledger through that boundary. Bounded autonomy and accountability for actions and outcomes must accompany any wider deployment.
MAS supplied no model evaluations, quantified economic estimates or deployment schedule. The announcement therefore offers an operating architecture rather than proof of safe performance.
Takeaway: MAS’s next test is whether banks can turn bounded mandates into working controls before agents move money beyond controlled pilots.
Source: unite.ai
More from James Calloway
Show full bio
News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.
791 articles