Mastercard Agent Pay for Machines: What It Is and Who It Is For

Short answer: Mastercard Agent Pay for Machines, launched June 10, 2026, is a multi-rail network service that lets AI agents and machines pay each other at machine speed across cards, bank accounts and stablecoins. Each agent gets a credential through Verifiable Intent, spending limits are checked before a transfer runs, off-chain cryptographic checks keep it fast, and Mastercard guarantees final settlement. It is built for high-frequency, low-value machine-to-machine payments, including microtransactions of a fraction of a cent. It is a rail and a network, not a buyer-side budget manager, so you still own the job of deciding what each of your agents is allowed to spend and recording it for finance.

Last updated July 2026.

What Mastercard announced

On June 10, 2026, Mastercard launched Agent Pay for Machines, a service designed to support secure, continuous machine payments across cards, accounts and stablecoins. The headline capability is machine-to-machine (M2M) payments: agents transacting with each other directly, at high velocity, executing chains of transactions that can include microtransactions worth a fraction of a cent. This is a step beyond consumer Agent Pay, which focuses on an agent paying a merchant on a person's behalf and which we put head to head with Visa's programme in Visa Intelligent Commerce vs Mastercard Agent Pay. Agent Pay for Machines is about the economy of agents paying agents.

More than 35 partners were named at launch, spanning card and payments firms such as Stripe, Adyen and Checkout.com and crypto-native names including Coinbase, OKX, the Solana Foundation, Polygon, RippleX and Skyfire. That breadth signals where the category is heading: a shared, multi-rail settlement fabric that both traditional card processors and stablecoin networks plug into.

How it works under the hood

Three pieces make the model work at machine speed. First, credentialing: each agent receives a credential through what Mastercard calls Verifiable Intent, so a payment carries proof of what the agent was permitted to do. Second, permissioning: the permissions a human grants an agent are logged to public blockchains, initially Polygon, Solana and Base among others, creating a tamper-evident record of authority. Third, speed: spending limits are checked before any transfer runs, using off-chain cryptographic checks so the decision does not wait on block confirmation, and Mastercard guarantees the final settlement.

The design goal is obvious once you see the use case. If two agents are settling thousands of sub-cent payments a second, you cannot route each one through a slow authorization path or a human. The network has to verify permission and enforce a limit in milliseconds, then settle reliably. That is what Agent Pay for Machines is engineered to do.

Who is this for?

Agent Pay for Machines is aimed at businesses building agent-to-agent economies: platforms where one agent buys a capability, a data feed, or compute from another agent and pays for it continuously. If your agents discover and purchase services from other agents, for example through an agent marketplace where capabilities are listed and priced, a machine-speed settlement rail is exactly what makes that practical. It is also relevant to anyone doing high-frequency metered billing, where each tool call or API request is a tiny charge that has to clear instantly.

It is less relevant if your agents mostly make ordinary purchases from human-facing merchants a few times an hour. For that pattern, a scoped virtual card on your existing issuer is simpler and gives you a cleaner reconciliation story. The M2M rail shines when volume and velocity are extreme.

What a network rail does not solve

Here is the part that matters for anyone deploying agents. A settlement network moves money and enforces the limit it was told to enforce. It does not decide, on your behalf, what each of your agents should be allowed to spend, why, and under what approval. Those are policy questions specific to your business, and they span every rail you use, not just one network.

Concretely, a network like Agent Pay for Machines can confirm an agent is permitted and settle the transfer. It is not the place where you set a 200 dollar daily cap on your research agent, require a human to approve anything over 500 dollars from your procurement agent, restrict a data-buying agent to three named vendors, or pull one shared, agent-attributed audit trail that covers card spend, ACH and stablecoin settlement together. That buyer-side control problem is yours regardless of which rail carries the money.

How Agentspay fits alongside it

Agentspay is the control plane that sits in front of whatever rails your agents use, including a network like this one. It enforces hard spend limits on amount, counterparty and velocity before money moves, routes anything above a threshold to a human for approval in Slack, email or a webhook, issues scoped single-use virtual cards by API, and writes an immutable audit trail attributed to the agent and its owner. Because it is rail-neutral, the same policy and the same audit trail apply whether a given payment settles on a card, over ACH, or as a stablecoin transfer through a machine-to-machine network.

The two are complementary. The network is the fast, guaranteed settlement layer. The control plane is where your rules live and where finance gets one clean record. You want both: a rail that clears at machine speed and a governance layer that makes sure the machines only spend what you actually authorized.

Frequently asked questions

When did Mastercard Agent Pay for Machines launch?

Mastercard launched Agent Pay for Machines on June 10, 2026. It is a multi-rail service for secure machine-to-machine payments across cards, bank accounts and stablecoins, built for high-frequency, low-value transactions including sub-cent microtransactions.

How is it different from consumer Agent Pay?

Consumer Agent Pay focuses on an AI agent paying a merchant on a person's behalf. Agent Pay for Machines focuses on agents paying each other directly at machine speed, supporting continuous high-velocity chains of transactions and microtransactions between machines rather than a single consumer checkout.

Which rails and chains does it use?

It settles across cards, bank accounts and stablecoins. Agent permissions are logged to public blockchains, initially Polygon, Solana and Base among others, and spending limits are verified with off-chain cryptographic checks before a transfer runs, with Mastercard guaranteeing final settlement.

Do I still need my own spend controls?

Yes. A settlement network enforces the limit it is given and moves money, but it does not decide what each of your agents should be allowed to spend or record it across every rail for finance. You still need a control plane that sets per-agent policy, gates high-value spend for human approval, and produces one unified audit trail.

The takeaway

Agent Pay for Machines is a real milestone: a major network turning on machine-to-machine payments across cards, accounts and stablecoins at machine speed. It solves settlement. It does not solve the buyer-side question of what your agents are allowed to spend and how you prove it later. That is why a rail-neutral control plane belongs above any settlement network. See how it works, compare the best agent payment platforms, or read our take on agent-to-agent payments.