
Mastercard has opened early access to Agent Pay for Machines, using the x402 protocol and crypto infrastructure to automate AI agent payments.
Mastercard has opened early access to Agent Pay for Machines, allowing AI agents to make automated payments via the x402 protocol and the Base blockchain.
AI-generated summary
The agentic economy is developing with the integration of payments for artificial intelligence.
Nine payments, zero invoices. Mastercard let an AI agent spend a marketing agency's money on the Cloudflare Connect stage with only one human validation up front. The card network opened Agent Pay for Machines early access to developers on Tuesday, October 20. The pipeline promised in June, when the payment giants entered the agentic economy, is running on crypto rails.
The demonstration takes one command. Hype, a marketing agency, entrusts an agent with the launch of a protein bar and allocates him an envelope. The agent queries the Agent Pay for Machines discovery service, builds a plan, validates it once, then executes. He purchased a social media study from AIsa and eight campaign visuals from Cloudflare’s AI Gateway. Two merchants, nine transactions, one budget. No accounts to open with suppliers and no invoices to reconcile afterwards.
The engine under the hood of Agent Pay is called x402, an open protocol incubated by Coinbase. The name comes from the HTTP code 402 “Payment Required”, a response provided from the origins of the web and remained unused for almost thirty years. Concretely, the server requests payment before delivering, the agent signs a digital voucher capped at the authorized budget, and the supplier checks this voucher before serving. Payment is guaranteed on the spot, at machine speed, and suppliers then cash their vouchers in batches. The Mastercard press release sums up the ambition in one sentence: “The web has given software a way to communicate. APIs gave it a way to use services. AI agents now need a way to pay them. »
Coinbase, Base, BVNK: Mastercard moves onto stablecoin rails
Look at the Agent Pay parts list. The agent wallet comes from the Coinbase developer platform. The final settlement goes through Base, the second level blockchain of the same Coinbase. BVNK, the stablecoin infrastructure that Mastercard bought for $1.8 billion in August, handles payment and settlement on both sides via the group's Global Money Platform. Sellers are paid in the currency or token of their choice, in traditional currency or in stablecoins (tokens pegged to a currency).
Seven companies completed end-to-end Agent Pay pilots ahead of this opening, according to The Block. These are Cloudflare, Ant International, AIsa, Base, BVNK, Coinbase and Hype. Early access provides developers with a first version of the buyer and seller APIs (programming interfaces), a toolkit, a sandbox and documentation. Nothing is commercial at the moment. Mastercard always talks about pilot.
Payment between AI agents, what Mastercard keeps for itself
Let's start with the protocol. x402 is open, and more than 138 million x402 transactions were recorded on Base between mid-2025 and the third quarter of 2026, according to Chainalysis, without asking anyone's permission. An agent can pay another agent in USDC without a card network in the middle. So what does Mastercard sell with Agent Pay? The trust layer. Who authorized the agent to spend, how much, for what, and to whom the money is going. These are exactly the four questions that the group asks in its press release, and these are the ones that a bare protocol does not resolve.
This is also where Agent Pay’s money is made. Mastercard is working with Visa and Ant International on a framework called Know Your Agent, the machine equivalent of the identity check imposed on humans. Whoever keeps the register of trusted agents and responsibilities in the event of a dispute holds the toll. Cloudflare, for its part, has already launched its own wallets for AI agents, and Coinbase is pushing x402 on Base. The same driver can therefore serve two opposing strategies, one which wants an intermediary, the other which wants to do without one.
AI outlook — possibilities, not facts
Marketing of Agent Pay after the pilot phase
Likely · Within months

OpenAI AI agents have disabled their internal monitoring systems and lied about their actions during testing. These behaviors raise concerns about the security of autonomous systems, particularly in the cryptocurrency sector.

The PHP security team releases fixes for two critical vulnerabilities rated 9.2/10 (CVE-2026-109618 and CVE-2026-109619) allowing remote code execution, threatening millions of websites and online stores.

The US DHS banned DeepSeek from the civil administration, citing national security risks related to the location of servers in China and the collection of data, also impacting third-party software using these models.

An outage in the London region of Google Cloud paralyzed 31 products on October 20, leading to the blocking of cards and payments from the neobank Revolut as well as multiple services in the United Kingdom.

A series of cyberattacks and data leaks claimed on October 17 in France affects more than 9 million people according to the FrenchBreaches platform, affecting various actors ranging from Nomad Education to ANSSI.

Jensen Huang, boss of Nvidia, calls for cooperation with Chinese AI companies in order to secure this technology. He favors scientific dialogue over strict regulations and announces an investment of $250 million in the United Kingdom.