Mastercard has opened early access to Agent Pay for Machines, a system designed to let artificial-intelligence agents find digital services, accept their terms and pay automatically within spending limits set by people. The company also demonstrated the system processing a live transaction at Cloudflare’s Connect conference.
The initiative addresses a developing challenge in the AI economy. Autonomous software can increasingly select tools and services for a task, but many online providers still require customers to create accounts, obtain API keys or establish separate billing arrangements. Those requirements can make it difficult for an agent to purchase services from providers it discovers during a job.
According to Decrypt, Mastercard first announced plans for Agent Pay for Machines in June. The company’s latest disclosure moves the project from an earlier concept toward developer access and a demonstrated payment process.
How Agent Pay works
Agent Pay for Machines is built on x402, an open protocol that enables online services to request payment. Mastercard’s system connects that payment process with its network and uses digitally signed “payment vouchers” to establish that an agent is authorised to spend within a specified limit.
The process begins when a person or business gives an AI agent an assignment and a spending cap. If the agent reaches a paid service, that service can respond with a “402 Payment Required” message. The agent then creates a signed voucher promising payment within the approved limit. The provider verifies the voucher before supplying the requested service.
The payment interactions take place off-chain at machine speed, while providers can collect and redeem multiple vouchers in batches later. In practical terms, the arrangement is intended to separate the agent’s authority to purchase from the need to create a new account or billing relationship for every service it uses.
The system does not give an agent unrestricted control over funds. Its activity is bounded by the terms and spending limit established by the human or organisation assigning the task.
What Mastercard demonstrated
At Cloudflare Connect, Mastercard’s Jorn Lambert showed a marketing agent working for the agency Hype. The agent was given a budget to plan a protein-bar advertising campaign and then searched for registered providers able to complete parts of the assignment.
The demonstration involved a two-step plan. One provider supplied social-media research, while Cloudflare supplied campaign images. The agent completed nine payments to two independent merchants under a single budget, without creating separate accounts or generating individual invoices for each transaction.
Several companies supported the payment flow. Coinbase’s developer platform provided the wallet, while Base, an Ethereum layer-2 network, supplied on-chain settlement. BVNK handled payment and settlement in either fiat currency or stablecoins through Mastercard’s Global Money Platform. Mastercard acquired BVNK in August, according to the source material.
Mastercard said the system follows end-to-end pilots involving Cloudflare, Ant International, AWS, BVNK and Coinbase. It is now making buyer- and seller-side application programming interfaces available, alongside a toolkit and a sandbox for developers.
Part of Mastercard’s stablecoin strategy
The project forms part of Mastercard’s broader effort to develop payment infrastructure for both stablecoins and AI agents. Stablecoins are digital tokens designed to maintain a stable value relative to a reference asset, commonly a fiat currency. In this system, they are one of the settlement options available through the Global Money Platform; the demonstration also supported fiat settlement.
Mastercard has recently detailed other initiatives in this area, including its Stablecoin Security risk tool and the Global Money Platform powered by BVNK. Agent Pay for Machines extends that work to transactions initiated by autonomous software rather than directly by a person at the point of payment.
The significance of the demonstration is that an agent completed a sequence of purchases across multiple merchants using one authorised budget. The early-access release gives developers a way to test the model, although the source material does not provide broader commercial rollout details or adoption figures.
Market picture
Moneyiar market data at the stated times on October 21, 2026, showed the following prices and moves:
- Cronos (CRO): $0.0583470745, up 3.6% over 24 hours and down 0.61% over seven days, as of 11:09 a.m. +03:30.
- Aster (ASTER): $0.7239085857, up 3.66% over 24 hours and 3.52% over seven days, as of 11:08 a.m. +03:30.
- Flare (FLR): $0.006433659, up 1.93% over 24 hours and down 5.9% over seven days, as of 11:09 a.m. +03:30.
- Ethereum Name Service (ENS): $5.6851195352, up 5.1% over 24 hours and down 4.49% over seven days, as of 11:09 a.m. +03:30.
