Mastercard and Visa Race to Control AI Agent Payment Rails as Banks Test Transaction Frameworks

The payment networks are moving fast to establish the infrastructure for AI agents to spend money on behalf of consumers, with DBS Bank and Westpac signing on as the first banking partners to pilot competing frameworks from Visa and Mastercard.

Singapore's DBS Bank became the first bank in Asia Pacific to test Visa Intelligent Commerce, the card network's API suite designed for what the industry calls "agentic payments"—transactions where AI agents autonomously complete purchases using credit and debit cards. Separately, Mastercard announced it had processed its first transaction in New Zealand using its Agent Pay framework, working with Westpac to purchase cinema tickets with full cardholder consent.

The pilots mark an inflection point for corporate treasury and payment operations teams. If AI agents become the primary interface for routine transactions—from travel bookings to supply purchases—finance leaders will need new controls, audit trails, and reconciliation processes that don't exist in today's card programs.

Both networks are essentially racing to answer the same question: when an AI agent swipes a card, who's actually liable, and how do you prove consent? The technical challenge is that traditional card networks weren't designed for non-human actors. Every participant in the payment chain—issuer, acquirer, merchant—needs to see that an agent conducted the transaction, not a fraudster spoofing one.

DBS and Visa have already completed real-world food and beverage transactions in Singapore, demonstrating that AI-powered agents can handle everyday purchases through what Visa describes as "secure, issuer-controlled flows." The phrase matters: it signals that banks, not tech companies, retain control over transaction authorization. The partners now plan to expand testing to online shopping and travel bookings.

The timing reflects Singapore's aggressive AI adoption. More than three-quarters of Singapore residents already use generative AI tools like chatbots in daily life, according to a Visa-commissioned study, and eight in ten Singapore consumers now rely on AI assistance for online shopping.

Mastercard's approach with Westpac appears similar in mechanics but different in branding. The cinema ticket purchase was "fully authorised with cardholder consent," according to the company, with every participant in the payment flow able to recognize an agent conducted the transaction. That visibility is critical—it's the difference between an authorized agent payment and what looks like account takeover fraud.

For CFOs, the immediate question isn't whether agentic payments will happen (they will), but how to prepare expense management and reconciliation systems for a world where AI agents book travel, order supplies, and pay invoices without human review of each transaction. The card networks are building the rails. Finance teams will need to build the guardrails.

The pilots also reveal an unspoken competition: both Visa and Mastercard are positioning themselves as the trusted infrastructure layer between banks and AI platforms. Whoever establishes the dominant framework for agent authentication and consent management effectively controls the payment rails for the next generation of commerce. Banks like DBS and Westpac are the proof points, but the real customers are the finance departments that will eventually need to audit what their AI agents bought last quarter.