Keynote Address by SMS Tan Kiat How at the Unveiling of StableX Agentic Network
17 November 2026
Good afternoon, Dr. Bo Bai, ladies and gentlemen.
I’m happy to be joining all of you today at the launch of the StableX Agentic Network.
We are entering a new phase of Artificial Intelligence (AI). For the past few years, much of our attention has been on what AI can tell us. Increasingly, AI can also act on our behalf — accessing systems, coordinating tasks and carrying out transactions.
This creates exciting possibilities for businesses and consumers. But it also raises an important question.
How do we enable AI agents from different organisations and platforms to interact and transact in ways that people and businesses can trust?
I believe this will be one of the defining challenges of the emerging agentic economy. And it is an area where Singapore can make a meaningful contribution.
Today, I would like to make three points.
First, trust has to be operational, not just conceptual. We need practical mechanisms to establish, enforce and continuously maintain trust. An agent should have a verifiable identity. We need to know who it represents and who is responsible for it.
It must also have a clear mandate. Knowing who an agent is does not automatically tell us what it is authorised to do.
An agent authorised to compare quotations should not automatically be authorised to make purchases. And an agent permitted to make purchases should not have unlimited authority to spend. These distinctions become critical when agents can take actions with real financial consequences.
We are already seeing useful progress here in Singapore.
Earlier this month, CSA, GovTech and IMDA issued a joint advisory on the safe and secure use of personal AI agents. It cautioned against granting general-purpose AI agents broad access to sensitive accounts, including banking, government and healthcare services.
The principle is straightforward: as AI agents become more capable, we must be deliberate about the authority and access we give them.
In financial services, MAS has issued AI Risk Management Guidelines, setting out supervisory expectations for financial institutions.
MAS is also working with financial institutions and AI firms on Safeguards for Agentic Finance at Runtime, or SAFR. SAFR explores how an agent's proposed actions can be evaluated before a transaction is executed, and how records can be maintained for accountability. This includes checking the agent’s identity and whether its proposed actions are authorised and within its mandate.
In Government, GovTech has developed a Whole-of-Government Agentic Risk & Capability Framework to help identify, assess and mitigate risks in agentic AI systems, with risk-based safeguards and measures.
It is also developing GovHarness, a shared infrastructure supporting the building and operation of AI agents, including evaluation, monitoring and secure connections to tools.
These efforts are just a start. It's encouraging that we're trying different things out. But I want to dive a bit deeper because these efforts reflect an important lesson.
A trustworthy model does not automatically make a trustworthy system. And a trustworthy agent does not automatically make a trustworthy interaction.
Trust must extend across the entire workflow. And it must be maintained over time.
An agent that is safe today may behave differently tomorrow as its model, tools, permissions or operating environment change. That is why continuous assurance is essential.
We need to monitor behaviour, detect unexpected actions, evaluate performance and retain the ability to intervene or revoke permissions. We must also learn from incidents and improve our systems as risks evolve.
Trust must be established before an agent acts, enforced while it acts, and continuously assured as it operates.
Second, trust must work across systems and borders. The agentic economy will not be built on a single model, platform or technology ecosystem.
Different businesses and sectors will have different needs. If every platform develops its own closed system for identifying and authorising agents, we risk creating a fragmented landscape where agents cannot easily interact.
That increases costs and complexity, especially for small and medium enterprises. Even for developed markets or advanced economies, small and medium enterprises face many barriers to engage in an agentic economy.
Every point of friction and every cost adds up for them. If you have a fragmented ecosystem, it will be prohibitive for many firms and companies to participate meaningfully in the agentic economy of the future. We therefore need open, interoperable approaches that allow different systems to establish trust.
I am heartened to see that the industry is already developing promising solutions. Visa's Trusted Agent Protocol seeks to help merchants distinguish legitimate AI agents from malicious bots. Google's Agent Payments Protocol explores how agents can initiate payments securely on behalf of users.
IMDA is also working with industry partners to explore whether existing internet infrastructure, including the Domain Name System, could support agent identification and verification across platforms.
These are still early days, and these approaches are still evolving. We should not assume that one technical architecture is necessarily the answer.
Instead, we should focus on the outcomes we want: reliable identity, verifiable mandates, enforceable safeguards and interoperability.
And there is a broader economic and strategic dimension. Interoperability creates value because different systems can work together. It preserves choice because businesses need not depend entirely on any single system.
For Singapore, this matters greatly. We benefit from being open and connected to different technology ecosystems.
Our resilience does not come from trying to build everything ourselves. It comes from retaining meaningful choices and developing the expertise to make those choices count.
Trust has to go beyond borders and bring different ecosystems together. Singapore has long played this role – whether through airports, or digital payment systems. In an agentic economy, we can play a similar role as well.
Third, Singapore wants to build these foundations together with industry. Singapore has experience building trusted digital infrastructure and bringing together regulators, technology companies and enterprises to solve difficult implementation problems.
But the agentic economy will not be built by any one government alone. Much of the trust infrastructure we need must be developed, tested and adopted by industry.
In this regard, I see three areas where we can collaborate.
First, interoperable Know Your Agent, or KYA, arrangements.
Organisations increasingly need to verify which agents they are dealing with, who stands behind them and what they are authorised to do.
But we should avoid a situation where every platform develops its own incompatible approach.
Industry can help establish common standards and verifiable credentials that allow agent identities and mandates to be checked across different organisations and systems.
Second, continuous assurance.
Verifying identity and authority at the outset is necessary, but insufficient.
Agents change. Their models are updated, their permissions evolve, and they encounter situations we may not have anticipated.
We need practical approaches to monitoring behaviour, detecting unexpected actions, sharing relevant assurance information and intervening when necessary.
This becomes especially challenging when agents interact across organisational boundaries.
Industry collaboration can help develop common testing methods, assurance signals and mechanisms for revocation and incident response — while respecting legitimate requirements for security, privacy and commercial confidentiality.
Third, trusted transactions in real-world environments.
We need to test how these arrangements work when agents from different organisations actually interact and transact.
Can they verify one another's authority?
Can safeguards be enforced before consequential actions?
Can we trace what happened, establish accountability and resolve problems when things go wrong?
Financial services, payments and commerce provide useful starting points.
This is why I welcome industry-led initiatives such as the StableX Agentic Network. MetaComp is exploring how agent identity, authorisation and compliance controls can support trusted transactions through its KYA framework.
I also welcome MetaComp's participation in the AI Verify Foundation, contributing to the wider work on testing, assurance and responsible AI deployment.
The real value of these efforts will be in demonstrating that trust can work across different systems — not just within one platform.
Government can support this work by convening partners, contributing technical and regulatory expertise, facilitating appropriate experimentation and helping establish common standards or infrastructure, where necessary.
Our collective objective should be to build the common foundations that allow different approaches to work together.
Let me conclude.
The next phase of AI will not simply be about making agents more intelligent. It will be about enabling agents to work together across organisations and systems, safely and confidently.
For Singapore, our ambition is to help build the trusted, interoperable foundations that make this possible through practical collaboration, experimentation and continuous learning.
The challenge is not simply to make agents interoperable. It is to make trust interoperable — and to keep that trust valid as agents act.
That is why initiatives such as StableX matter. They allow industry to develop and test the mechanisms that can make trusted agentic transactions possible.
I congratulate MetaComp and its partners on today's launch. I look forward to working together to build the foundations for a trusted agentic economy of the future.
Thank you.
