The scenario sounds like science fiction until you think about how much of it already exists. An AI assistant renews a subscription, pays for a cloud service it uses on your behalf, splits a bill with another assistant that booked a flight, and settles the whole thing without a human touching a card. To make that work at scale, someone has to agree on how machines pay each other. Visa, Stripe, and Google have joined an open-source project called AIPay Protocol to build that agreement.
The project aims to define a standard protocol for payments initiated by AI agents, software that acts on a user’s behalf. The founding members want agents to be able to hold credentials, authorize charges, and settle transactions the way a person would, but faster and without manual steps, according to people familiar with the project. The goal is not a new payment network but a common language that existing networks, banks, and merchants can all speak.
The use cases are already emerging. Subscription services are the simplest: an assistant notices a renewal, checks the budget, and pays. Model-to-model payments come next: when one AI service calls another’s application programming interface, the caller pays per use, automatically. The most complex case is multi-agent commerce, where several agents cooperate on a task, booking travel, procuring supplies, managing a supply chain, and need to settle among themselves. In each case, the money must flow through the existing, regulated payment system, and the protocol is designed to keep it there.
That last point explains why Visa and Stripe are involved. The companies make their living moving money, and they would rather machines use their rails than a new, unregulated alternative. The protocol does not create a new currency or a new settlement network; it standardizes how agents request, authorize, and confirm payments on top of the rails that already exist. For Visa, that means more transactions. For Stripe, it means the next generation of its business is an application programming interface call, which is what Stripe has always been.
Google’s interest is broader. The company’s assistants and its cloud business both depend on AI agents becoming a normal part of the economy, and a payment standard removes a bottleneck that would otherwise slow adoption. Google has also been building its payments infrastructure for years, and it sees agent payments as an extension of that work. The three companies bring different strengths to the project, and their agreement on a shared standard is itself notable, since they compete in adjacent markets.
The project is open source, which is both a virtue and a risk. An open standard is more likely to be adopted widely, but it also means anyone can implement it, including actors who may not follow the rules the founding members intend. The governance of the project, who approves changes and who sets the security requirements, will determine whether the standard is trusted or bypassed, analysts said.
Security is the obvious challenge. A payment protocol designed for machines must solve problems that human payments already struggle with: proving identity, preventing fraud, and assigning liability. If an agent charges $10,000 by mistake, or is tricked into doing so, the question of who is responsible has no easy answer. The founding members are working on authorization models that keep the human in control of what agents can spend, with limits, approvals, and audit trails built into the protocol, according to people familiar with the matter.
Regulators are watching from the sidelines, for now. Payments are among the most regulated parts of the economy, and a standard that lets software spend money will attract scrutiny, particularly around consumer protection and money laundering. The project’s backers say the protocol is designed to work within existing rules rather than around them, a position that will be tested the first time an agent’s payment is disputed.
The commercial stakes are large. Forecasters expect agent-driven commerce to grow into a significant share of digital payments over the next several years, and the companies that define the standard will hold an advantage in the market that follows. That is why the project is being built in the open rather than as a proprietary product: the founders are competing to be the rails, and the fastest way to win is to make everyone use your specification.
For the rest of the industry, the protocol is a signal that the machine economy is coming sooner than expected. The plumbing for AI payments will be standardized, regulated, and competitive, just like the plumbing for human payments. The only difference is that the customer will be a piece of software with its own budget, and the first question it asks will be the same one every customer asks: how much does this cost?
