Innovation & Tech

MAS drafts 'know your agent' protocol as AI agents move into banking

MAS managing director Chia Der Jiun said AI agents handling payments could threaten financial stability. The Singapore regulator is drafting a "know your agent" protocol and an Agentic Money Sandbox.

By Amara Osei

4 min read

MAS flags financial stability risks as AI agents gain access to financial services
MAS flags financial stability risks as AI agents gain access to financial servicesfotoblend / pixabay

What's News

  • MAS managing director Chia Der Jiun warned on Thursday at the Singapore FinTech Festival 2026 that AI agents moving money could threaten financial stability.
  • SAFR's first phase set out four principles for safe agentic finance: bounded autonomy, clear identification and authority, specific mandates, and proportionate controls.
  • MAS's next phase develops a "know your agent" protocol covering an agent's identity, authority, mandate and origin assurance.
  • MAS will establish an Agentic Money Sandbox with the Future of Finance Institute to test AI agents with tokenized deposits and stablecoins.
  • Chia cautioned consumers against giving personal AI agents their financial account login credentials, citing insufficient assurance on safe access.

Monetary Authority of Singapore managing director Chia Der Jiun warned on Thursday that AI agents handling payments, fund allocation and financial decisions could threaten financial stability. He spoke at the Singapore FinTech Festival 2026.

Chia said consumers and businesses must keep meaningful control over the objectives, mandates and actions of their AI agents, and that autonomy has to come with clear accountability. "As agents move into activities with financial consequences, authority, accountability and control become essential," he said.

The regulator's concern is operational rather than abstract. Financial institutions need a way to establish that an agent is authorized by a customer and is acting inside its mandate.

Someone also has to answer when an agent executes an incorrect transaction or acts beyond the instructions given to it. Chia cautioned consumers against handing personal AI agents their financial account login credentials, saying there was insufficient assurance such access could be provided safely.

MAS is working with the financial industry on a trusted protocol and secure channel through which agents can reach financial services. The work sits inside its Safeguards for Agentic Finance at Runtime program, or SAFR.

Lessons from SAFR's first phase produced four principles for safe agentic finance, according to Chia: bounded autonomy, clear identification and authority, specific mandates, and proportionate controls. Agents should have enough discretion to be useful while staying within clearly defined boundaries. Mandates should specify permissible actions, limits and conditions, and an agent's identity and authority must be established.

What must banks prove about an agent?

In the next phase, MAS will facilitate a "know your agent" protocol for trusted information exchange between customer agents and financial institutions. It will cover an agent's identity and whom it represents, the authority granted to it, the mandate it operates under, and the assurance surrounding its origin and reliability.

That would let institutions set the terms and scope of access, and establish secure channels for agent transactions.

MAS will also establish an Agentic Money Sandbox with the Future of Finance Institute to explore AI agents and programmable money in corporate treasury management and payment services.

In treasury, agents could monitor liquidity positions and upcoming payments, decide when and where funds should transfer, and execute permitted transactions using tokenized deposits.

For payment services, banks and payment service providers could allow customer agents to execute payment instructions using digital money, including tokenized deposits and stablecoins. Participating institutions would verify agents' identities and authority before applying appropriate risk controls, Chia said.

What changes when agents transact with each other?

Chia said agents could cut the cost and time involved in searching, comparing, negotiating and executing transactions, producing more frequent and faster consumer and business decisions.

Consumers could use agents to compare products, prices and terms across multiple providers, negotiate on their behalf and monitor price movements to secure better deals. Businesses could automate procurement and inventory management, reach a wider pool of suppliers, and continuously compare prices, delivery times and contractual terms.

The added speed and frequency could introduce new risks to market competition and pricing behaviour, Chia warned.

He noted that emerging studies of agent interactions in simulated environments had identified the possibility of agent collusion, even without explicit instructions to collude. Such behaviour could result in higher prices or weaker competition.

More capable agents could also secure better prices and terms than less capable ones, creating unequal access to economic opportunities depending on the quality of available AI models.

Agent-to-agent interactions could eventually give rise to dynamic markets in which prices and commercial terms are continuously determined through negotiations between autonomous systems.

Financial institutions may need to strengthen their value propositions to both customers and AI agents, while considering whether transaction limits are necessary to manage risk from increasingly autonomous financial activity.

Regulators, Chia said, will need to consider the implications of AI agents' participation in financial markets and their ability to move money at greater speed.

MAS is working with the financial industry to develop common standards, governance practices and infrastructure. He said regulatory clarity, clearly defined governance responsibilities and industry collaboration would help the sector adopt transformative technologies while maintaining resilience.

Ultimately, MAS aims for an interoperable ecosystem of trusted interactions among AI agents, financial institutions and digital money, with meaningful human control and accountability central to the agentic economy. The test ahead is whether banks impose agent transaction limits before autonomous negotiations start setting prices.

Source: technode.global

Share this article:

More from Amara Osei

Amara Osei

Show full bio

Senior reporter covering companies, deals and strategy in entrepreneurship, startups and venture funding at Business Bearings.

2,422 articles

Related articles

« Previous articleNext article »