Skip to main content
AI Banking SecurityMAS Singapore· 7 min read· in Business

Singapore Readies 'Know Your Agent' Protocol to Regulate Banking AI

Singapore's central bank has initiated plans to develop a standardized 'Know Your Agent' verification architecture alongside a live testing sandbox for autonomous software in commercial banking. The framework establishes authenticated access channels to prevent users from sharing login passwords with artificial intelligence assistants.

By Bo Feng

Share this story

Listen

On November 19, 2026, on the Festival Stage at the Singapore EXPO, Monetary Authority of Singapore managing director Chia Der Jiun stepped to the lectern to address more than 70,000 attendees gathered for the Singapore FinTech Festival. He announced that the central bank is formulating a formal 'Know Your Agent' protocol alongside a dedicated regulatory sandbox to govern autonomous software transacting across the banking system.[1][2][6]

The policy initiative marks the world's first comprehensive central bank framework designed specifically to govern autonomous artificial intelligence agents that execute payments and reallocate funds. Rather than prohibiting algorithmic financial assistants, Singapore's primary monetary regulator is establishing an identity and verification standard analogous to traditional Know Your Customer banking rules.[2][3][6]

Speaking before executives from 840 central banks and global regulatory bodies, Chia cautioned that the rapid adoption of consumer and enterprise artificial intelligence has outpaced the digital plumbing of modern commercial finance. While generative models are increasingly trusted to draft correspondence and analyze balance sheets, financial infrastructure was built strictly around direct human authentication.[1][6]

The central bank chief highlighted that an increasing number of early adopters have begun granting automated personal assistants direct login credentials to retail bank accounts and corporate treasury interfaces. That shortcut, Chia warned, bypasses critical audit trails and exposes account holders to catastrophic operational, legal, and fraud vulnerabilities.[2][3]

"Some users are giving agents their login credentials to financial accounts," Chia stated during his keynote address. "I would not advise giving personal agents such access, as there is insufficient assurance of safety."[2][6]

The Security Flaw in Shared Passwords

Under existing retail and commercial banking protocols, a financial institution receiving an automated instruction cannot verify whether the account holder approved the specific transfer. When an artificial intelligence agent logs into an online banking portal using a customer's personal username and password, the bank sees only standard customer activity rather than an external automated script.[2][3][4]

This arrangement breaks down completely whenever an automated tool encounters ambiguous instructions, suffers prompt injection, or executes a hallucinated transaction. If an autonomous agent rebalances a corporate liquidity pool or purchases goods beyond its intended parameters, existing commercial law offers no clear answer regarding who bears financial liability.[1][3]

The five identity and authorization pillars established under Singapore's proposed Know Your Agent protocol.

"The financial institution does not know if this is an agent that is acting on the full authorisation and control of the customer, nor whether the agent is transacting within its mandate and instructions," Chia explained. "If agents transact incorrectly or beyond their instruction, who bears responsibility?"[2][6]

The dilemma has already prompted major international institutions to construct defensive barriers against unverified bots. In October 2026, Charles Schwab Corporation confirmed it was actively blocking third-party artificial intelligence agents from accessing client accounts, underscoring the growing friction between autonomous technology and legacy institutional security.[5]

Singapore's central bank argues that outright bans cannot survive the broader transition toward automated commerce. Industry projections from Gartner and McKinsey estimate that business-to-business transactions conducted by software agents could reach $15 trillion globally by 2030, while consumer retail agent spending could exceed $1 trillion over the same period.[1][5]

Five Pillars of Agent Verification

To resolve the impasse, the Monetary Authority of Singapore is launching the next operational phase of its Safeguards for Agentic Finance at Runtime program, known as SAFR. SAFR establishes five standardized credentials that an artificial intelligence agent must present to a financial institution before initiating any monetary transfer.[1][3][6]

Under the proposed Know Your Agent protocol, banks will evaluate an agent's technical identity, the specific legal principal it represents, and the explicit scope of delegated authority granted to it. Financial systems will also inspect the operational mandate governing permissible transaction thresholds, alongside cryptographic assurance regarding the software's provenance and developer reliability.[2][3][6]

This structured exchange replaces insecure screen-scraping and password-sharing with dedicated, authenticated application programming interfaces. Financial institutions will be able to verify an agent's mandate in real time, granting calibrated access that expands or contracts based on the specific risk profile of the requested transaction.[1][2][4]

"The greater the discretion and impact of action given an agent, the stronger the assurance and controls needed," Chia noted. "This will allow financial institutions to determine the terms and scope permitted to an agent, and establish a dedicated secure channel for agent access."[1][6]

By formalizing these verification channels, Singapore seeks to convert autonomous software into recognizable economic actors rather than anonymous web traffic. The protocol ensures that every algorithmic instruction carries an unalterable audit log, clarifying whether an erroneous trade originated from user misconfiguration, software defect, or bank processing failure.[3][4]

Projected agentic commerce volume by 2030 compared against early central bank tokenization baselines.

Testing Liquidity in the Agentic Sandbox

To pressure-test these standards in a live environment, the monetary authority has established an Agentic Money Sandbox in partnership with the Future of Finance Institute, a collaborative body founded on November 3, 2026. The sandbox unites commercial lenders, technology developers, and central bank supervisors to study agent behavior across two initial use cases.[1][3][6]

The primary corporate use case focuses on automated treasury management, allowing corporate software agents to continuously monitor multi-bank liquidity positions and upcoming debt maturities. Under predetermined company mandates, the agent can autonomously allocate funds between yield accounts and settle supplier invoices using tokenized commercial bank deposits.[1][4][6]

Participating lenders retain ultimate gating authority over each transaction, enforcing strict real-time risk controls and velocity caps on all outgoing transfers. Corporations define the boundaries of permitted operations, ensuring that the artificial intelligence cannot deploy capital outside pre-approved asset classes or exceed daily expenditure limits.[1][3]

The second sandbox track evaluates consumer payment workflows, testing how autonomous agents initiate retail purchases using regulated stablecoins and tokenized bank deposits. Payment service providers in the pilot verify agent identity and permissions at runtime, testing multi-signature approvals for transactions that exceed routine micro-payment thresholds.[1][4]

The sandbox runs in tandem with Singapore's broader digital currency infrastructure, which saw the central bank natively issue 100 million Singapore dollars of four-week tokenized MAS Bills on November 6, 2026. Settled in wholesale central bank digital currency across five primary dealers, the sovereign ledger provides the high-speed settlement foundation required by autonomous agents.[4][6]

Market Velocity and Financial Stability

While autonomous systems promise to eliminate administrative drag and transaction friction, the central bank warned that hyper-efficient software introduces systemic risks to macroeconomic stability. In traditional banking, deposit movement is constrained by manual processing, human evaluation, and standard business operating hours.[5][6]

Software agents operating on continuous loops can detect basis-point yield differentials across competing institutions in milliseconds. Central bank officials warned that automated algorithms could instantaneously reallocate billions of dollars across platforms in pursuit of marginal returns, dramatically accelerating the speed and severity of liquidity strains.[5][6]

Structure of the Agentic Money Sandbox testing autonomous liquidity management and programmable payment flows.

"Agents have the potential to move money and allocate assets more quickly for smaller incentives," Chia warned. "Financial institutions will have to strengthen their value proposition to customers and agents, and potentially consider whether limits are needed on agent-instructed transactions to manage risk."[5][6]

The monetary authority also pointed to emerging academic research showing that autonomous agents operating in financial markets can inadvertently develop coordinated pricing strategies. Even without explicit algorithmic instructions to collude, interacting machine-learning models can learn that tacit cooperation maximizes joint margins, potentially undermining competition and harming consumer welfare.[1][6]

"Autonomy has to be bounded," Chia stated to conclude his keynote address. "Actions and outcomes have to come with accountability."[1][5][6]

Roadmap to 2027 Institutional Deployment

Rather than imposing premature statutory rules, Singapore's regulatory roadmap relies on iterative industry collaboration throughout 2027. The Singapore Dollar Test Network is being transitioned into an operational public utility named the SG Digital Ledger, establishing a unified institutional rail for tokenized settlement.[4][6]

Rather than imposing premature statutory rules, Singapore's regulatory roadmap relies on iterative industry collaboration throughout 2027.

The five primary dealers that cleared the initial 100 million Singapore dollar tokenized debt issuance—DBS, OCBC, UOB, HSBC, and Standard Chartered—are expanding operational testing to incorporate foreign government securities and money-market fund collateral into the programmable ledger architecture.[4][6]

Central bank supervisors in Singapore plan to publish technical specifications for the Know Your Agent standard following initial results from the Future of Finance Institute sandbox. The resulting protocols will determine whether financial institutions worldwide can safely open their vaults to autonomous software or remain locked behind manual human authentication.[1][3][6]

Key points

  1. The Monetary Authority of Singapore launched a 'Know Your Agent' protocol to verify artificial intelligence tools transacting with commercial banks.
  2. Central bank chief Chia Der Jiun warned consumers and businesses to stop sharing personal bank account login credentials with autonomous software.
  3. A new Agentic Money Sandbox with the Future of Finance Institute will test AI treasury management and programmable tokenized deposit payments.
  4. Regulators warned that high-speed automated rebalancing could accelerate bank liquidity risks, necessitating strict transaction velocity limits.

What we don’t know

  • What specific technical standard or cryptographic protocol MAS will adopt to ensure cross-border compatibility with foreign central banks and financial jurisdictions.
  • How civil courts and financial dispute resolution bodies will apportion legal liability when an autonomous agent executes a mistaken transfer caused by model hallucination.
  • Whether commercial retail banks will set hard limits on daily transaction values initiated by personal artificial intelligence assistants.

How we got here

  1. Nov 2024

    Singapore expands Project Guardian and Project Orchid to test tokenized assets, digital currency settlement, and wholesale interbank ledgers.

  2. Oct 2026

    Major brokerage firms including Charles Schwab begin blocking external AI agents from accessing client financial accounts over safety concerns.

  3. Nov 3, 2026

    The Future of Finance Institute begins operations in Singapore to study institutional AI adoption, agent deployment, and financial risk governance.

  4. Nov 6, 2026

    MAS issues S$100 million of four-week tokenized MAS Bills natively on the Singapore Dollar Test Network settled in wholesale CBDC.

  5. Nov 19, 2026

    MAS managing director Chia Der Jiun announces the Know Your Agent protocol and the Agentic Money Sandbox at the Singapore FinTech Festival.

Central Bank Regulators 40%Fintech and Banking Developers 35%Risk and Security Analysts 25%
Central Bank Regulators
Argues that autonomous financial agents require bounded autonomy, cryptographic verification, and strict institutional limits to safeguard market stability.
Fintech and Banking Developers
Advocates for standardized programmable APIs and tokenized deposit infrastructure that allow autonomous agents to operate efficiently without relying on brittle screen-scraping.
Risk and Security Analysts
Warns that autonomous agentic transactions risk accelerating liquidity drains and opaque liability failures if software agents are granted direct bank access before rigorous runtime controls are proven.

Perspectives this story doesn't cover

  • Consumer advocacy groups evaluating user legal liability when an autonomous agent makes a mistaken purchase.
  • Open-source AI developer foundations advocating for lightweight agent interoperability protocols without proprietary banking lock-in.

Sources

Source coverage

6 outlets

3 viewpoints surfaced

Central Bank Regulators 40%Fintech and Banking Developers 35%Risk and Security Analysts 25%
  1. [1]Singapore Business ReviewCentral Bank Regulators

    MAS explores AI agent protocol and tokenisation framework

    Read on Singapore Business Review →
  2. [2]Fintech News SingaporeFintech and Banking Developers

    MAS Chief Warns Against Giving AI Agents Your Bank Login

    Read on Fintech News Singapore →
  3. [3]i10xRisk and Security Analysts

    Know Your Agent: Singapore's MAS Plans a Protocol for AI Agents in Banking and Warns Users Not to Share Their Logins

    Read on i10x →
  4. [4]The DefiantFintech and Banking Developers

    MAS Plans SG Digital Ledger and 'Know Your Agent' Protocol

    Read on The Defiant →
  5. [5]EmergentRisk and Security Analysts

    MAS Warns AI Agents Could Move Money Faster, Raise Risks

    Read on Emergent →
  6. [6]Monetary Authority of SingaporeCentral Bank Regulators

    Advancing the Future of Finance: From Foundation to Scale

    Read on Monetary Authority of Singapore →
Listen

Comments

Stay informed

Every angle. Every day.

Get Business stories with full source coverage and perspective breakdowns, free every day.