AI Agents Can Pay Each Other. Mastercard Wants the Toll

If AI agents can pay each other at machine speed, who decides what they’re allowed to buy before the money moves?

Mastercard has launched Agent Pay for Machines, a service that lets AI agents and machines complete transactions among themselves, according to PYMNTS. The product extends Mastercard’s earlier Agent Pay program into a more demanding corner of commerce: high-frequency, low-latency, low-value payments initiated by software rather than a person tapping a button.

That’s the real shift. Agentic commerce is moving from “recommend this product” to “complete this task and pay for what’s needed.” Mastercard wants its network to handle the trust layer before agent-to-agent payments become a patchwork of private databases, crypto rails, platform credits, and custom integrations.

Why should businesses care that Mastercard is letting AI agents pay each other?

Because many agent transactions won’t look like normal checkout.

Mastercard says Agent Pay for Machines is built for payments that are “continuous, embedded, permissioned and executed at machine speed.” It can support credentialing, controls, and guaranteed settlement across cards, stablecoins, and other payment types.

That combination matters. A human purchase usually has a visible moment of consent: a checkout page, a saved card confirmation, a banking app approval, or a card-on-file charge. Machine-led payments can happen inside a workflow where the commercial action is only one step in a longer automated task.

Mastercard’s own example is straightforward: a merchant’s AI agent could launch a store’s web presence by buying a domain name, a hosting service, images, and checkout pages.

That is not science fiction. It is a tightly scoped business process with multiple vendors and several small payments. The hard part is not imagining the task. The hard part is proving that the agent had permission, stayed inside its limits, and paid the right counterparties.

“Agent Pay for Machines will create the conditions for a superbloom of AI business models,” Mastercard Chief Product Officer Jorn Lambert said. “Machine payments can make it possible for services to be bought and sold among agents at fundamentally different scales than payments today, very high volumes, very small values, very fast and at extremely low latency.”

Analysis: Mastercard is not just adding a new checkout option. It is trying to define how money moves when software becomes the buyer.

How does Mastercard Agent Pay for Machines differ from regular digital payments?

Agent Pay for Machines is designed for transactions where the payer may be an AI agent or connected machine, not a human sitting in front of a checkout screen.

That changes the payment problem.

Mastercard introduced Agent Pay in April 2025 as an agentic AI-driven payments program meant to integrate payment experiences into generative AI-powered conversational platforms. Agent Pay for Machines pushes the same idea into machine-to-machine activity.

The difference is volume and tempo. A person may authorize a few purchases in a session. An agent could trigger many smaller payments as part of a task. Fortune reported one possible use case: an AI agent accessing data piecemeal from a website and paying as it goes.

That is why “low-value” matters. A payment that is too small for an invoice can still need controls, records, and settlement certainty.

Which agent-to-agent payments are actually described so far?

The clearest Mastercard example is the merchant web launch scenario.

An AI agent working for a merchant could buy:

• Domain name: The web address for the store.
• Hosting service: The infrastructure to run it.
• Images: Commercial assets for the storefront.
• Checkout pages: The payment-facing layer of the site.

That example is useful because it shows the type of transaction Mastercard is targeting: several small purchases, each tied to a business objective, executed as part of one automated workflow.

Fortune added another concrete example: an AI agent paying for data access in small increments. That points to a different class of machine payment, where the value is not one large purchase but many tiny permissioned interactions.

The business logic is clear. If agentic commerce becomes real, many payments will sit below the threshold where manual approval makes sense but above the threshold where companies can ignore controls.

This is where the story connects to the broader shift toward payments embedded inside software. XOOMAR has covered how product teams already face rail selection and control problems in Embedded Finance Platforms Can Make or Break Your Launch. Agent-to-agent payments raise the stakes because the buyer may be autonomous within a defined scope.

How can AI agents spend money safely without giving machines a blank check?

The safety layer has to answer three questions:

• Identity: Which agent initiated the payment?
• Authority: What did the human or business allow it to do?
• Outcome: Did the transaction match the approved instruction?

Mastercard has already introduced one piece of that control stack. In March, the company unveiled Verifiable Intent, an open-source, standards-based framework for agentic commerce. Mastercard said it is designed to link a consumer’s identity, specific instructions, and transaction outcome into a single tamper-resistant record.

That matters when something goes wrong. The record creates a cryptographic audit trail that parties can consult if a dispute arises.

Fortune reported that Mastercard’s new protocol stores human-granted permissions for AI agents on a blockchain rather than a private database, making that information available to multiple parties that need to verify whether an agent acted as instructed. A Mastercard spokesperson told Fortune the company initially chose to log those permissions onto Polygon, Solana, and Base, among others.

The minimum control set for agent payments is easy to state and hard to implement:

• Spending caps: How much can the agent spend?
• Vendor limits: Who is it allowed to pay?
• Purpose restrictions: What task is the payment tied to?
• Time windows: How long does permission last?
• Escalation rules: When does a human need to approve?

Mastercard says Agent Pay for Machines can support credentialing, controls, and guaranteed settlement. That is the pitch: automated commerce without removing the payment network’s familiar trust machinery.

The lesson from other recurring and automated payment categories is that reliability becomes part of the product. As XOOMAR wrote in Failed Payments Crown Subscription Payment Gateways, payment failure is not just a back-office issue when the transaction is tied directly to user access or business continuity.

Why are payment networks moving before AI-agent volumes prove the market?

Because the interface to commerce may change before the payment rails do.

Fortune reported that agentic payment volumes are still a fraction of broader commercial flows. Mastercard’s own product chief did not frame this as an immediate revenue windfall.

“Am I expecting that this is going to be a huge revenue driver for Mastercard next year? No,” Lambert told Fortune. “Do I think it’ll be a meaningful new addressable market for us over the next five years? I think so.”

That is a sober quote. It also explains the timing.

Mastercard is collaborating with more than 30 initial partners to validate priority use cases, establish common rules, and accelerate adoption across industries. Fortune named Adyen, Coinbase, and Cloudflare among companies working with Mastercard on the protocol.

Adyen’s role is especially relevant because merchant acceptance will decide whether agentic payments remain a lab project or become a usable payment pattern.

“Building these foundations with partners like Mastercard, openly and with merchant outcomes at the center, is how we ensure this next era of commerce works for everyone in the ecosystem,” said Karan Katyal, head of agentic commerce at Adyen.

Analysis: the race is not only about processing payments. It is about setting the rules for agent identity, permission, dispute evidence, and settlement before fragmented standards harden.

What has to happen before Mastercard’s AI-agent payments become mainstream?

The launch answers one question and opens several harder ones.

Mastercard has shown the direction: agents and machines can be credentialed, constrained, and connected to settlement rails that include cards and stablecoins. It has also put Verifiable Intent into the story, which gives the model an audit trail rather than relying only on trust in the agent.

But adoption will depend on unresolved practical details:

• Common rules: Mastercard says partners are working to establish them, but the market has not yet seen how uniform they will be.
• Enterprise controls: Businesses will need narrow approval workflows before they give agents payment authority.
• Merchant acceptance: Agent payments need counterparties that can recognize the agent and process the transaction cleanly.
• Liability clarity: If an agent follows flawed instructions or pays the wrong party, dispute handling will matter.
• Production proof: Pilots and partner announcements are not the same as daily transaction volume.

Companies should not read this as a signal to hand broad budgets to autonomous agents. The practical first step is smaller: identify tasks where the payment amount is low, the vendor set is known, the business rule is clear, and the audit trail is mandatory.

Mastercard’s Agent Pay for Machines is an early building block for machine-to-machine commerce. The next test is whether businesses trust agents enough to let them pay repeatedly, not just recommend once.

Disclaimer: This XOOMAR analysis is for informational and educational purposes only. It is not financial, investment, legal, tax, or professional advice. It does not provide buy, sell, hold, price-target, portfolio, or personalized recommendations. Verify information independently and consult qualified professionals before making decisions.

Impact Analysis

• Mastercard is positioning its network as a trust layer for agent-to-agent commerce.
• Businesses may soon automate small, frequent payments across vendors without manual checkout steps.
• Permissioning and settlement controls will be critical as AI agents start spending money autonomously.