Singapore is moving to address one of the biggest challenges emerging from artificial intelligence: how to trust software that can make decisions, access digital systems and carry out tasks on behalf of people and businesses.
The issue was brought into focus in Parliament on Wednesday, November 4, when Senior Minister of State for Digital Development and Information Jasmin Lau outlined the Government’s efforts to develop safeguards for increasingly autonomous artificial intelligence agents.
Her remarks followed a proposal from Marine Parade-Braddell Heights MP Tin Pei Ling for a national digital identity infrastructure that would allow different organisations to verify an AI agent’s identity, ownership and authorised responsibilities.
The proposal comes as AI technology moves beyond conventional chatbots.
Traditional AI applications are primarily used to generate information, answer questions and make recommendations. AI agents, by contrast, can interact with other systems, use digital tools and execute multiple steps to complete tasks with varying levels of human supervision.
That creates new opportunities for businesses and consumers — but also introduces risks involving unauthorised transactions, access to sensitive information, cybersecurity and accountability.
Lau said trust in AI agents depends on three fundamental questions:
- Who is the agent?
- What is it authorised to do?
- Who is accountable for its actions?
These questions could become increasingly important as AI agents begin handling more consequential activities, including financial transactions and access to private or sensitive data.
The central challenge for Singapore is no longer simply developing more capable AI.
It is ensuring that people can identify, control and hold someone accountable for the technology acting in their name.
Tin Pei Ling proposes a national digital identity system for AI agents
The parliamentary discussion was prompted by Tin Pei Ling’s proposal for a national digital identity infrastructure designed specifically for AI agents.
She suggested establishing a registry that would allow different parties to verify an AI agent’s identity and the mandate under which it operates.
The idea is comparable in principle to the way digital identity systems help organisations establish who is accessing a service.
However, an AI agent’s identity would need to establish more than the software’s name or platform.
It would also need to show which individual, company or organisation is responsible for deploying it and what authority the agent has been granted.
For example, an AI agent authorised to search for flight tickets should not automatically have permission to purchase a ticket, access unrelated personal information or transfer money.
Similarly, an agent deployed by a business to manage customer enquiries should not be able to change financial records or access confidential employee information without appropriate authorisation.
Tin also proposed that significant changes to an agent’s mandate should require fresh, documented authorisation.
Her proposal included the idea of biometric passporting for AI agents, alongside open and interoperable standards that would allow different platforms and organisations to recognise verified agents.
The objective is to prevent a fragmented environment in which every company creates its own identity system, making it difficult to verify agents operating across different services.
A common infrastructure could help establish trust between platforms without requiring organisations to rely entirely on separate verification arrangements.
However, a national registry alone would not prevent every form of misuse. The system would still need reliable authentication, permission controls, monitoring and ways to respond when an agent behaves outside its authorised boundaries.
Jasmin Lau says Singapore supports the proposal’s objectives — but a national registry is not yet confirmed
Responding to Tin in Parliament, Lau said the Government shares the ambition of making Singapore a place where individuals and businesses can use AI agents with confidence.
She said the Government wants agents to operate securely across different platforms and organisations without creating separate digital environments or closed systems.
But Lau did not announce that Singapore would immediately establish a national AI-agent registry.
Instead, the Government intends to work with industry, build on existing infrastructure and emerging international standards, and identify gaps that may require additional solutions.
The approach is important because several initiatives are already underway.
Rather than creating an entirely new system before understanding how AI agents are being deployed, Singapore wants to examine how existing identity, payment and internet infrastructure could support agent verification.
This leaves open the possibility of a national registry in the future, but its necessity and design still need to be assessed.
Singapore is already testing ways to verify AI agents before they handle money
One of the most consequential areas for AI-agent adoption is financial services.
Imagine an AI agent instructed to purchase an item, pay a bill or complete a transaction on behalf of its user.
The merchant or payment provider would need to know whether the agent is legitimate, whether the person behind it has authorised the transaction and whether the requested payment falls within the permitted limits.
Without reliable verification, businesses could face difficulties distinguishing authorised AI activity from impersonation, fraudulent instructions or compromised software.
Lau pointed to work already underway in the private e-commerce and payments sectors, including initiatives involving Visa and Google.
In financial services, the Monetary Authority of Singapore is developing Safeguards for Agentic Finance at Runtime, working with financial institutions and AI companies.
The system is intended to verify an AI agent’s identity and authority before transactions are executed.
It would also maintain auditable records of the agent’s actions.
That record-keeping is particularly important when something goes wrong.
If an AI agent makes an unauthorised payment or performs a transaction outside its mandate, investigators and financial institutions need information that can help establish what happened, what permissions were given and which parties were responsible.
The wider objective is to make AI-powered financial activity more secure without preventing legitimate automation.
WWC One Media previously examined the growing competition to establish trusted AI payments in its report, Visa, Mastercard and Ant Want AI Agents to Spend Your Money — But One Question Could Decide Whether Anyone Trusts Them.
The question is becoming increasingly relevant as AI moves from recommending purchases to potentially completing them.
Singapore is exploring whether the internet itself can identify AI agents
Another initiative involves the internet’s Domain Name System, commonly known as DNS.
DNS is the infrastructure that helps translate website domain names into the addresses needed for computers and other devices to communicate.
Lau said the Infocomm Media Development Authority is working with GoDaddy, Infoblox, Cloudflare and Google to explore whether DNS can be adapted to help identify and verify AI agents across different platforms.
The potential significance is considerable.
If agents are going to interact with websites, online services and other digital systems, organisations need ways to establish whether those agents are legitimate and what they are permitted to access.
Adapting existing internet infrastructure could help make verification more interoperable.
However, the initiative remains exploratory. It should not be interpreted as a completed global identification system or a guarantee that all AI agents will become automatically trustworthy.
Identity verification is only one part of the problem.
A legitimate agent can still make mistakes, misunderstand instructions or be manipulated into performing actions its user never intended.
That is why identity needs to be combined with permission management, monitoring and accountability.
GovTech is developing an internal registry for government AI agents
Singapore’s public sector is also preparing for wider use of agentic AI.
Lau said the Government Technology Agency of Singapore, or GovTech, is developing an internal registry to help government agencies track the AI agents they deploy and the public officers responsible for them.
The initiative addresses an important governance issue.
When a government agency uses multiple AI agents across different departments, it needs to know which systems are operating, what information they can access and who is responsible for overseeing them.
An internal registry could help agencies maintain visibility over their AI deployments.
It could also support accountability when an agent performs an unexpected action or accesses information beyond its intended purpose.
For example, an agent designed to assist with administrative work should not automatically be given unrestricted access to every government database.
Its permissions should reflect the actual task it is expected to perform.
This approach is consistent with Singapore’s wider effort to apply risk-based controls to AI systems.
Singapore has already identified cybersecurity and privacy risks in AI agents
The latest parliamentary discussion did not emerge in isolation.
Singapore has been examining the risks of increasingly autonomous AI systems through testing initiatives involving government agencies and technology companies.
In May 2026, the Cyber Security Agency of Singapore published findings from its AI Agents Sandbox, conducted with Google, GovTech and IMDA.
The project examined how computer-using AI agents could operate in practical environments.
It explored three areas:
- Automated quality assurance: Testing government digital services and identifying problems such as inactive pages and incorrect website information.
- AI safety testing: Using agents to help evaluate chatbot systems across different languages and formats.
- Social assistance applications: Exploring how AI could guide applicants and social workers through complex application processes.
The trials demonstrated potential productivity benefits, but they also highlighted important risks.
These included the need for human oversight, stronger cybersecurity, privacy protection and controls over how agents access information and use digital tools.
One particularly important concern was indirect prompt injection.
This occurs when an AI agent encounters malicious instructions embedded in content it processes and is manipulated into taking actions that were not intended by its user.
For example, an agent asked to summarise a webpage could encounter hidden instructions designed to make it disclose information or perform an unrelated action.
The risk becomes more serious when the agent has access to email accounts, databases, payment services or other sensitive systems.
The sandbox’s findings reinforce why AI agents need carefully defined permissions and safeguards that can be tested in realistic conditions.
The official findings are available through the Cyber Security Agency’s report on the AI Agents Sandbox.
Singapore is considering stronger rules for high-risk AI systems
The Government’s approach is also developing beyond the question of identity.
In October, Minister for Digital Development and Information Josephine Teo said Singapore was studying whether stronger safeguards were needed for high-risk AI applications.
Potential measures could include more rigorous testing, independently verifiable evidence that safety measures work, tighter deployment controls and stronger regulatory oversight.
Teo also said the Government takes a risk-based approach to public-sector AI deployment.
Before an agent is used, agencies consider which systems and data it can access, what actions it is authorised to perform and the consequences of unintended or unauthorised activity.
She said there had been no reported cyberattack against Singapore government agencies involving an unsupervised AI agent at the time of her parliamentary response.
That clarification does not mean the risks have disappeared.
It means the Government is examining potential weaknesses before they become serious incidents.
CNA reported that Singapore is also considering how existing cybersecurity incident-reporting channels could better identify and respond to incidents involving AI.
WWC One Media covered the same policy direction in its report, Singapore Reports No Rogue AI Attack on Government Agencies — But Stronger Safeguards Are Coming.
The broader challenge is ensuring that safeguards remain effective as AI systems gain more independence.
AI agents are becoming more capable — and the consequences of mistakes are changing
The distinction between a chatbot and an AI agent is becoming increasingly important.
A conventional chatbot might provide instructions for booking a hotel.
An AI agent could potentially search for accommodation, compare prices, enter personal information, make a reservation and complete payment, depending on the permissions it receives.
That additional capability can save time and simplify everyday tasks.
But it also creates more opportunities for mistakes.
An agent might misunderstand a user’s instructions, select the wrong service, disclose information to an unintended recipient or perform an action that exceeds its authority.
Cybersecurity risks are another concern.
The more tools an agent can access, the more opportunities there may be for malicious instructions or compromised systems to influence its behaviour.
This is why experts and policymakers increasingly argue that AI agents should operate with limited permissions, clear boundaries and meaningful human oversight.
WWC One Media has also examined the wider cybersecurity implications in its report, INTERPOL Warns Agentic AI Could Automate Cyberattacks as Criminals Scale Scams at Machine Speed.
The concern is not simply that AI can help people carry out cyberattacks.
It is that increasingly autonomous systems could make certain malicious activities faster and easier to scale.
Singapore wants international standards before AI agents operate across borders
AI agents are unlikely to remain within the boundaries of one country or company.
A single agent could interact with a payment provider in one country, a merchant in another and a cloud platform operated elsewhere.
That creates a difficult question: how can different systems recognise and trust an agent when they follow different identity and security standards?
Lau said Singapore is looking at international developments, including work in China, the United States and Estonia.
Singapore also intends to strengthen regional capabilities in AI governance, testing and assurance when it assumes the ASEAN chairmanship in 2027.
The country wants to support the adoption of international standards and common testing tools for agentic AI.
This could help establish more consistent expectations for organisations deploying agents across different markets.
Without interoperability, businesses could face a fragmented system in which an agent recognised by one platform cannot be verified by another.
But international cooperation will take time.
Countries may differ in their approaches to data protection, digital identity, cybersecurity, consumer protection and legal responsibility.
Singapore’s challenge will be to support common standards while ensuring that its own safeguards remain effective.
The biggest unanswered question: Who takes responsibility when an AI agent causes harm?
The technology may be capable of acting independently, but the question of accountability remains a human and organisational issue.
If an AI agent makes an unauthorised payment, discloses confidential information or changes an important record, responsibility cannot simply disappear behind the software.
Potentially relevant parties could include the organisation deploying the agent, the company providing the technology, the person who authorised its actions and other parties involved in operating the system.
The precise allocation of responsibility will depend on the circumstances, applicable laws and the evidence.
A registry could help identify the organisation or individual responsible for an agent, but it would not automatically determine legal liability in every incident.
That is why identity, authorisation, audit records and human oversight must work together.
Lau’s closing message captured the issue: agents may act on behalf of people, but someone must remain answerable for what they do.
The Bottom Line
Singapore is preparing for a future in which artificial intelligence will not simply answer questions — it will increasingly act on behalf of people, businesses and government agencies.
Tin Pei Ling’s proposal for a national digital identity infrastructure has brought an important issue into the spotlight: how can organisations verify an AI agent’s identity, establish its authority and know who is responsible when something goes wrong?
Jasmin Lau said the Government supports the objectives behind the proposal, but a national registry has not yet been confirmed. Existing initiatives involving financial transactions, internet infrastructure and GovTech’s internal agent registry will help inform the next steps.
The stakes extend beyond technology companies.
AI agents could eventually handle purchases, payments, administrative work and access to sensitive information. If they are given greater freedom to act, the systems controlling their permissions and accountability will become increasingly important.
Singapore wants to encourage AI adoption while ensuring that trust does not depend on promises alone.
But as AI agents become capable of spending money, accessing private information and making decisions on our behalf, will Singapore’s safeguards be ready before the technology moves faster than the rules designed to control it?
