Singapore’s monetary authority launches AI agent sandbox for corporate cash management
Singapore’s central bank is building a controlled testing ground for autonomous AI agents to manage corporate cash and payments using tokenized deposits, with banks retaining full authority over execution. This marks the first regulatory framework to explicitly govern agent access to financial accounts and addresses a critical institutional risk: how to allow AI automation while preventing unauthorized or erroneous transactions.
- MAS and the Future of Finance Institute will launch the Agentic Money Sandbox, allowing AI agents to execute treasury tasks within company-set mandates and bank-controlled limits.
- The SAFR protocol will give banks four data points: agent identity, who it represents, its authority level, and assurance of its origin before granting account access.
- DBS, OCBC, Standard Chartered and UOB are launch partners; MAS has already issued S$100 million in tokenized bills settled on Singapore dollar wholesale CBDC.
- Nov 19 MAS Managing Director announcement of Agentic Money Sandbox at Singapore FinTech Festival
- S$100M tokenized MAS Bills issued and settled on SGD wholesale CBDC this month
- 4 data points SAFR protocol will require banks to obtain before granting agent access to accounts
Singapore’s Monetary Authority announced on November 19 (Thursday) a regulatory sandbox for artificial intelligence agents that autonomously manage corporate treasury and payments using tokenized deposits, according to reporting by Cryptopolitan. The move addresses a gap in financial regulation: existing rules govern human account access but offer no framework for autonomous software agents making financial decisions. MAS Managing Director Chia Der Jiun unveiled the Agentic Money Sandbox at the Singapore FinTech Festival, where the central bank outlined how banks and corporate clients can test agent-driven cash management within predefined boundaries before live deployment.
MAS Creates Four-Factor Verification for AI Agent Account Access
The Agentic Money Sandbox will require participating banks to use a protocol called SAFR (Safeguards for Agentic Finance at Runtime) to authenticate AI agents before granting them access to corporate accounts.
Under SAFR, banks will obtain four pieces of information: the agent’s identity and the entity it represents, the specific authority it holds, the mandate under which it operates, and assurances about the agent’s origin and reliability. This layered approach mirrors know-your-customer rules applied to human employees, but adapted for software that can execute transactions at machine speed.
Chia warned against a common practice already emerging in retail markets: users giving agents their banking login credentials directly. “I would not advise giving personal agents such access, as there is insufficient assurance of safety,” he said, flagging that liability rules remain unsettled if an agent executes transactions outside its mandate or makes errors.
Businesses see benefits for agents to manage cash, payments and liquidity.
Chia Der Jiun, MAS Managing Director
Four Major Singapore Banks Join Initial Testing Phase
DBS, OCBC, Standard Chartered and UOB will serve as launch partners for the sandbox. In the treasury use case, an AI agent authorized by a company would monitor liquidity positions and upcoming payment obligations, then recommend or execute fund transfers across tokenized deposit accounts held at partner banks.
The company sets the agent’s overall mandate; the bank sets hard limits on transaction size, frequency and recipient. Before executing any instruction, the agent must signal what it intends to do, allowing the bank’s risk systems to apply controls and either approve or block the action.
On the payments side, customer agents would initiate transfers using digital money such as tokenized deposits or stablecoins. Banks and payment providers would verify the agent’s identity and authority before settlement, applying the same risk controls used for high-value wire transfers or automated clearing house payments.
Singapore Dollar Tokenized Infrastructure Now Supports Live Settlements
Singapore has already moved beyond testing to live issuance of tokenized central bank money.
MAS issued S$100 million of tokenized MAS Bills this month and settled them on the Singapore dollar wholesale central bank digital currency (CBDC), known as SGD Testnet. This infrastructure will transition into a working utility called SG Digital Ledger.
Under the MAS BLOOM initiative, StraitsX, OCBC and Standard Chartered successfully completed US dollar payments across two separate tokenized deposit networks in minutes, demonstrating that cross-network settlement is now technically feasible at scale.
Visa, Mastercard and Ant International are separately developing a Know-Your-Agent interoperability framework through BuildFin.ai, a platform launched by MAS, to allow card networks, wallets and agent platforms to identify and trust agents across different systems.
The CCS read. We see the regulatory template here: MAS is not banning agents from financial accounts but defining the technical and governance standards that make them safe enough for institutional adoption. Banks retain execution authority; companies retain mandate authority; the agent operates only within that envelope. This model could become the blueprint for other jurisdictions, shifting the question from whether AI touches institutional cash to how tightly it must be monitored.
Watch for the first corporate treasury client to go live on the sandbox and for the liability framework MAS publishes; Chia’s question about who bears the cost of agent errors remains unresolved, and that answer will determine whether large corporates treat agents as a convenience or a material risk.
From the CCS interview archive
All 1,000+ interviews
Mark Zalan on GoMining's non-custodial Bitcoin payments for merchantsMark Zalan, GoMining · Jun 2026
Q Protocol on implementing decentralized governance across blockchainsNicolas Biagosch, Q Protocol · Sep 2024
ZIGChain expands wealth generation strategies for Web3 usersAbdul Rafay Gadit, ZIGChain · Jan 2025


