Machine Payments Protocol

In computing, a protocol is simply an agreed-upon set of rules by which two programs communicate with each other. A Machine Payments Protocol is such a rule set for money: it defines how one program pays an amount to another. What makes it special is that no human presses a payment button in the process. The paying program is given prior authorization with limits, say, ten euros per day, and handles the rest on its own. This is why it is also referred to as machine-based or autonomous payment. Such rule sets are currently being developed, and no single standard has yet become established.

Why software suddenly needs its own wallet

Classic payment methods are built for humans. You type in a card number, confirm in a banking app, click Buy. Every one of these steps assumes that someone is sitting in front of a screen. A program working alone at night gets nowhere with that.

That is exactly what is now becoming a problem. AI assistants are taking on tasks in multiple steps: they search databases, call up maps, have texts translated. Many of these services cost money, often tiny amounts per call. At a tenth of a cent, a credit card transaction isn’t worth it, because the fee alone amounts to many times that sum.

A Machine Payments Protocol is meant to close this gap. It makes amounts payable that are too small for conventional systems. At the same time, it promises control: whoever lets their program make payments wants to be able to specify exactly what it may pay for and up to what sum. Without these limits, a malfunctioning assistant would be a financial risk.

The path from program to paid service

The process resembles a toll booth. A program requests a service, say, a weather database. The service does not immediately respond with data, but with a notice: this information costs a certain amount, here is the payment address. The requesting program then sends the payment and makes the same request again, this time with proof of payment. Only then does the service deliver.

For the program to be able to pay at all, it needs a kind of wallet, in technical jargon called a wallet. It contains no cash, but a digital key that authorizes payments. Many approaches use stablecoins for this, that is, digital coins whose value is firmly pegged to a real currency such as the dollar. The reason is practical: such transfers happen within seconds and cost almost nothing, even for tiny amounts.

The most important part is the limitation. The owner sets rules in advance, such as a daily budget or a list of approved providers. A common misconception is that such a protocol gives the AI free access to a bank account. In fact, it works more like a prepaid card with a tight limit. Once the balance is empty, that’s it.

Who is working on it and what it means for users

The term has been appearing regularly in the news since 2025. Coinbase released x402, an open approach for payments directly within the web protocol. Google introduced its own variant with the Agent Payments Protocol, in which card providers such as Mastercard are also involved. OpenAI and Stripe are also working on ways for an assistant to make purchases on a user’s behalf.

In everyday life, little of this is noticeable for now. It becomes visible first among developers who call data services at a rate of several per second. The next step will be assistants that not only put together a trip but also buy the train ticket themselves right away. Whether and when this happens depends less on the technology than on questions of liability and on regulation by supervisory authorities.

What remains open above all is who is liable when a program buys something incorrectly. With a credit card, you can dispute a charge. With an automatic payment made in seconds, that is more difficult. That is why protocols, spending limits, and traceable receipts are a fixed part of every serious approach.