Singapore Backs Industry Trust Checks for AI Agents That Pay
Singapore says AI agents that buy and pay need verifiable identity, a clear mandate and continuous checks. MetaComp launched an industry network to test this, but the government gave no rules, dates or funding.
By Utopia Data & AI Team6 min read
In brief
- On 17 November 2026, Singapore's Senior Minister of State Tan Kiat How unveiled the StableX Agentic Network, an industry-led initiative launched by MetaComp, and said trust for AI agents must be "established before an agent acts, enforced while it acts, and continuously assured as it operates".[1]
- He proposed three areas for industry work: interoperable Know Your Agent arrangements, continuous assurance, and trusted transactions in real settings, starting with financial services, payments and commerce.[1]
- The speech sets no new rule, deadline or funding. It points to existing work by the Cyber Security Agency, GovTech, IMDA and the Monetary Authority of Singapore.[1][2]
What Singapore said on 17 November
Singapore wants the companies building AI agents to agree on how those agents prove who they are. On 17 November 2026, Senior Minister of State Tan Kiat How gave a keynote at the unveiling of the StableX Agentic Network, and the Ministry of Digital Development and Information (MDDI) published the text. The page was updated on 18 November.[1]
The speech calls the network an industry-led initiative that lets companies "develop and test the mechanisms" for trusted agentic transactions. It congratulates MetaComp "and its partners" on the launch. The published text does not state where the event took place, and it gives no figures, partner list or launch timetable for the network itself.[1]
Trust before, during and after an agent acts
The core of the speech is a three-part test. The minister said an agent needs a verifiable identity, a clear mandate and continuous assurance. In his words, "Trust must be established before an agent acts, enforced while it acts, and continuously assured as it operates." The text treats trust as something that has to stay valid over time, not a one-off approval at sign-up.[1]
He also warned against assuming that a good model makes a safe system. "A trustworthy model does not automatically make a trustworthy system," the text reads. That line shifts attention from the AI model to the wiring around it: the credentials an agent carries, the limits on what it can spend, and the logs that show what it did.[1]
Three areas where industry is asked to work
The speech proposes three areas of collaboration. The first is interoperable Know Your Agent arrangements, meaning shared ways to identify an agent and the person or firm behind it. The second is continuous assurance, which means checking an agent while it runs. The third is trusted transactions in real-world environments, tested with live partners rather than in a lab.[1]
The minister named financial services, payments and commerce as "useful starting points". He argued that fragmented, closed systems would raise costs, and that small and medium enterprises would feel this most. The text asks for open, interoperable approaches that focus on outcomes rather than one technical design. It says that "the challenge is not simply to make agents interoperable. It is to make trust interoperable."[1]
Government work already underway
The speech lists work by several agencies. The Cyber Security Agency of Singapore, GovTech and IMDA issued a joint advisory earlier in November on the safe use of personal AI agents. The advisory page is dated 6 November 2026 and says it covers the safe and secure use of such agents. We could not open its full PDF, so we do not report its detailed guidance.[1][2]
The Monetary Authority of Singapore has issued AI Risk Management Guidelines, according to the speech, and is working on SAFR, short for Safeguards for Agentic Finance at Runtime, with financial institutions and AI firms. GovTech has built a Whole-of-Government Agentic Risk and Capability Framework and is developing GovHarness. IMDA is working with industry on whether the Domain Name System could help identify and verify agents. The text also cites Visa's Trusted Agent Protocol and Google's Agent Payments Protocol as industry efforts.[1]
What MetaComp says its framework does
MetaComp, a company licensed by the Monetary Authority of Singapore as a Major Payment Institution, describes its StableX Know Your Agent (KYA) framework on its own site. The page says it extends the know-your-customer and know-your-business approach from people and companies to AI agents. It has four pillars: agent identity and registration, authority and permission control, behaviour monitoring and risk intelligence, and ecosystem and interaction governance.[3]
The page says MetaComp introduced the framework on 21 April 2026 at Money20/20 Asia in Bangkok, and publishes it for adoption by financial institutions, regulators and network partners. It names no adopting institution. Its claims to be the first such architecture are MetaComp's own, and we found no independent review. The speech says MetaComp is exploring how agent identity, authorisation and compliance controls can support trusted transactions through KYA.[3][1]
Why Malaysia should care
Our analysis: Malaysian merchants, fintech firms and banks that sell to Singapore will meet agent-initiated payments sooner than local rules arrive. The speech sets no Malaysian requirement. It does give Malaysian SMEs a checklist of questions to ask any vendor that lets an AI agent spend money: who is the agent, what may it do, and who answers when it errs.
Malaysian SMEs that sell or pay online
AI agents may soon place orders and payments for buyers. The speech says closed, fragmented systems would cost SMEs more, so open identity checks matter to small sellers.[1]
Practical move: Ask each payment or commerce vendor how it identifies an AI agent and who is liable when an agent errs.
Banks, fintech firms and payment providers
Singapore treats finance, payments and commerce as the first test beds. The speech lists work by MAS, GovTech and IMDA, but no obligation yet applies to firms outside Singapore.[1]
Practical move: Map which current customer checks could extend to agents, and watch MAS guidance for cross-border effects.
What Malaysians can do now
- Read the MDDI speech and the CSA, GovTech and IMDA advisory page in full.
- List every AI agent in your business that can spend money or place orders.
- Give each agent a written limit, a named human owner and an action log.
What we still do not know
What we still do not know
- Which firms beyond MetaComp take part in the StableX Agentic Network, and when testing begins.
- Whether Singapore will turn any of these ideas into rules, and whether Malaysian regulators will follow.
- How the network handles liability, data protection and cross-border use of agent identities.
Sources
- Keynote Address by SMS Tan Kiat How at the Unveiling of StableX Agentic NetworkMinistry of Digital Development and Information, Singapore, 17 November 2026
- Joint Advisory on the safe and secure use of personal AI agents (AD-2026-015)Cyber Security Agency of Singapore, 6 November 2026
- StableX Know Your Agent (KYA) FrameworkMetaComp, 21 April 2026
- File:Skyline, Marina Bay, Singapur, 2023-08-16, DD 77-79 HDR.jpgWikimedia Commons, 16 August 2023
- File:Singapore FinTech Festival.jpgWikimedia Commons, 15 November 2017
- File:Raffles Place.jpgWikimedia Commons, 19 September 2008
Facts come from the MDDI text of Senior Minister of State Tan Kiat How's keynote of 17 November 2026, the CSA advisory page dated 6 November 2026, and MetaComp's own KYA page, all read on 18 November. MetaComp's descriptions and first-of-its-kind claims are its own and we did not verify them. We could not open the advisory PDF and did not test the network. Sections marked as our analysis are Utopia's interpretation. The three photographs are licensed contextual images of Singapore and none shows MetaComp or the network. No generated image is used.
Last fact check: 18 November 2026. Factual correction? devteam.utopia@gmail.com.


