Mastercard and Visa Move on AI Payments

The strange part of this payments story is that the biggest card networks are not starting with faster checkout or lower fees. They are starting with a trust problem: how do you let a software agent buy something without handing it your actual card number or unlimited spending power. (mastercard.com) Mastercard said on April 6 that it has already rolled out authenticated agentic transactions across multiple Association of Southeast Asian Nations markets, with early live activity in Singapore and Malaysia and a regional artificial intelligence Centre of Excellence planned for Singapore later in 2026. United Overseas Bank is the first banking partner named in the rollout. (mastercard.com) (electronicpaymentsinternational.com) Mastercard’s pitch is built on tokenisation, which is the payments version of swapping your house key for a single-use valet key. Instead of passing around the real card credentials, the system uses tokenised credentials and what Mastercard calls Agentic Tokens so an artificial intelligence agent can pay without exposing the original account details. (mastercard.com) (electronicpaymentsinternational.com) The second piece is “verifiable intent,” which is Mastercard’s way of proving the user actually meant for the agent to make that purchase. The company pairs that with payment passkeys and end-to-end auditability so the bank, merchant, and card network can trace what the user approved and what the agent executed. (mastercard.com) Visa is taking a parallel route, but with a broader developer platform. Visa says its Intelligent Commerce system lets artificial intelligence agents shop and buy inside the Visa network, and on April 8 it added Intelligent Commerce Connect as a single integration layer for secure agent payment initiation, tokenisation, spend controls, and authentication. (visa.com) (cointelegraph.com) The clearest example came from Nevermined, which said on April 9 that it integrated Visa Intelligent Commerce with Coinbase’s x402 protocol so agents can buy digital goods and services autonomously under cardholder-defined guardrails. In plain English, Visa supplies the card rails, Coinbase supplies an internet-native payment protocol, and Nevermined is trying to turn that into software that can actually complete a purchase. (cryptobriefing.com) (financialit.net) That “guardrails” language is the whole game. If a human shopper taps “buy,” fraud systems mostly ask whether the card is real and the device looks familiar; if an agent shops for you, the system also has to ask whether the task, merchant, amount, and timing match the permission you gave the agent in the first place. (mastercard.com) (visa.com) That is why both companies keep talking about authorisation rather than just automation. A payment only works if the merchant accepts it, the bank approves it, and the fraud stack does not choke on behavior that looks “non-human,” so the commercial prize is not just new volume but higher approval rates on transactions started by software. (qz.com) (visa.com) Southeast Asia is a useful test bed for Mastercard because it combines fast digital commerce growth with lots of cross-border payment complexity across 10 countries. If agent-led buying works there with a bank like United Overseas Bank, Mastercard gets a live demonstration that the controls can survive different markets, currencies, and merchant setups. (mastercard.com) (developingtelecoms.com) The near-term use case is not a robot going on a shopping spree. It is narrower jobs like renewing a subscription, buying a data feed, booking a routine service, or completing a purchase after the user sets a budget and merchant rules, which is why Nevermined is focused on publishers and digital merchants first. (financialit.net) (cryptobriefing.com) If these systems work, checkout changes from “enter your card” to “delegate a task.” The winners will not be the companies with the flashiest chatbot, but the ones that can prove, transaction by transaction, that an artificial intelligence agent acted inside a user’s exact instructions and still got the payment approved. (mastercard.com) (visa.com)