AI agents have moved from payment-industry presentations into limited live testing in Southeast Asia. Mastercard says authenticated agentic transactions have been completed in Singapore and Malaysia, and Visa has enrolled banks and payment companies from ten Asia-Pacific markets in a controlled issuer-readiness programme. For regional finance teams, that is evidence that authorisation and identity controls are being tested on real payment rails; not evidence that autonomous systems already lower cross-border costs.
What is live in Southeast Asia
Mastercard says its ASEAN transactions use tokenised credentials, verifiable intent and end-to-end auditability. Its model is designed to preserve a tamper-resistant record of what a user authorised when an AI agent acts. UOB participated in the initial regional testing, alongside local banks in individual markets. International Monetary Fund
Visa’s Agentic Ready programme takes a different but related step. Issuers can test how agents initiate and complete transactions in a controlled, production-grade environment using identity, token, risk and control functions. The programme is live in Singapore, Malaysia, Thailand and Vietnam as well as six other Asia-Pacific markets, with regional banks including DBS, GXS, OCBC, UOB, Maybank, CIMB, KBank, Techcombank and VPBank listed among early participants.
Why control layers matter more than autonomy
These deployments show that an agent can be identified, constrained and tied to a user mandate. They do not yet show broad commercial adoption, unattended treasury authority or a uniform cross-border operating model. Mastercard’s announcement refers to a first wave and further expansion; Visa describes issuer testing and validation rather than a general production rollout for businesses.
What finance teams should demand next
The International Monetary Fund’s April note explains why that distinction matters. It separates agentic payment systems into intent and orchestration, authorisation and control, and settlement. Adaptive AI can help interpret an objective or choose among routes, but payment execution still needs deterministic rules, legal certainty, audit trails and clear responsibility.
For cross-border payments, the IMF identifies possible uses in exchange-rate monitoring, liquidity planning, compliance and route selection. Those capabilities could reduce manual work or improve timing, but the note also highlights authorisation traceability, opacity, cybersecurity, correlated behaviour and unresolved liability. It does not provide measured Southeast Asian savings from live deployments.
Source note
A practical adoption test for regional businesses is therefore narrower than the promise of autonomous payments. Buyers should ask who can set the mandate, what amount and counterparty limits apply, how an agent is identified, when human approval is required, which record proves intent, and who carries liability when an instruction or model is wrong. Until pilots publish comparable cost, speed, exception and loss data, agentic payments should be treated as a control-and-integration programme rather than a proven cost-reduction product.