Google has introduced the Agent Payments Protocol (AP2), an open standard designed to let AI agents complete secure payments across different platforms. The move points to a practical question facing agent-driven commerce: how can a transaction be trusted when software is acting on a user’s behalf?
What AP2 Is Trying To Solve
AI agents are increasingly being discussed not just as tools that answer questions, but as systems that can take actions. Payments are one of the most sensitive actions they could perform, because a purchase must be authorized, traceable, and understood by every party involved.
AP2 is meant to provide a shared framework for that process. Instead of each platform creating its own separate way for agents to pay, the protocol aims to support secure payment activity across different services.
The source describes AP2 as an open standard. That matters because agent payments involve more than one participant: the user, the agent, the merchant or platform, and the payment provider all need a way to understand what was approved and why.
Built On Agent2Agent
AP2 builds on the existing Agent2Agent protocol. In plain terms, that places the payment layer on top of an earlier effort focused on agent-to-agent interaction.
That relationship is important because payments are rarely isolated. An AI agent may need to communicate with another system before a transaction can happen. A standard for payment can therefore work more smoothly when it connects to a standard for agent communication.
The protocol also supports multiple payment methods. The source names credit cards, stablecoins, and bank transfers. That range suggests AP2 is not being framed around one payment rail, but around a broader model for agent-driven commerce.
Digital Mandates Put User Intent At The Center
A core part of AP2 is the use of digital mandates. These are cryptographic authorizations that lock in user intent.
That phrase is the key to understanding the protocol. If an AI agent makes a payment, the system needs a reliable way to show that the action matches what the user approved. A digital mandate is intended to make that approval verifiable.
The source says these mandates are meant to keep transactions verifiable and secure in two situations:
- Real-time purchases, where the user is directly involved.
- Automated transactions, where the user is not directly involved at the moment of payment.
The second case is especially important for AI agents. If a user sets conditions in advance, an agent may later act without the user watching every step. AP2’s design centers on preserving proof of intent even when the purchase is automated.
Why Backing From Payment And Software Companies Matters
The initiative is already backed by more than 60 companies. The source names Mastercard, PayPal, Coinbase, and Adobe among them.
That kind of backing is relevant because payment standards need broad participation to be useful. A protocol for AI agent payments becomes more valuable when it can be recognized by different platforms, payment providers, and software companies.
The goal, as described in the source, is to create a unified framework for a standardized and trustworthy system for agent-driven commerce. In practice, that means reducing ambiguity around who authorized a payment, which method was used, and how the transaction can be verified.
For users, the central issue is trust. For businesses, it is interoperability. For payment providers, it is whether agent-initiated transactions can fit into systems that already handle sensitive financial activity.
Where The Protocol Goes Next
Google has made the documentation available on GitHub. That gives developers and companies a place to examine the protocol and consider how it could fit into their own systems.
AP2 does not remove the need for careful implementation. The protocol sets out a framework, but the value of that framework depends on how consistently it is adopted and how clearly digital mandates are handled across real transactions.
Still, the direction is clear from the source: Google wants AI agent payments to move toward a common, verifiable model rather than a patchwork of platform-specific approaches. As AI agents take on more transactional roles, the ability to prove user intent may become one of the defining requirements for trustworthy automated commerce.