MoonPay’s PayBox Gives ChatGPT and Claude the Power to Make Payments By Matt Brienen Published: 2026-07-29T23:23:32+00:00 MoonPay is moving artificial intelligence beyond recommendations and into real-world transactions with PayBox, a new wallet designed for AI assistants such as ChatGPT and Claude. The product allows users to fund a wallet and then instruct an AI assistant, using ordinary language, to complete supported purchases on their behalf. That could include shopping online, booking a flight, reserving a restaurant table, or paying an internet service that accepts transactions from AI agents. It is an important step in the development of agentic commerce. AI assistants have become increasingly capable of researching products, comparing prices, and planning purchases, but they have generally stopped before the final transaction. PayBox is intended to close that gap by giving an assistant controlled access to money. --- ### How PayBox Works PayBox functions as a payment vault and non-custodial wallet that can be connected to compatible AI platforms. Users can supply funds through a bank account or cryptocurrency and then ask their assistant to perform a transaction. For example, a user might ask Claude to find a suitable flight within a particular budget and book it through an agent-compatible service. The same basic process could be used to pay for a restaurant reservation or purchase an item from a supported online merchant. MoonPay says the experience does not require users to write code or install specialized software. Claude users can add PayBox through the platform’s Connectors menu, making the service more accessible than developer-oriented tools that require command-line configuration. However, the wallet cannot automatically transact with every website. A merchant or service must support the infrastructure needed to interact with AI agents before PayBox can complete a payment there. --- ### Users Can Place Limits on Their AI Giving an AI assistant access to money immediately raises questions about control, mistakes, and unauthorized spending. PayBox attempts to address those concerns with configurable guardrails. A user can establish a spending ceiling, such as allowing purchases below $100, or require notification and approval before a payment is made. This creates a model in which the AI can research and prepare a transaction while the user decides how much independence it receives. Those safeguards will be central to the product’s credibility. A convenient payment experience will mean little if users cannot confidently review purchases, stop an incorrect transaction, or resolve disputes involving fraud and merchant errors. The broader agentic-commerce industry is still working through those issues. Traditional card networks have spent decades building chargeback and consumer-protection systems, while autonomous digital wallets introduce new questions about who is responsible when an agent misunderstands a request. --- ### Open Payment Rails Are the Crypto Connection PayBox is built around x402, an open payment protocol derived from the HTTP 402 “Payment Required” status code. The technology is designed to let software and AI agents discover a price, submit payment, and receive a digital service through an internet-native process. MoonPay is positioning this open design as an alternative to payment systems confined to a single company’s platform. The wallet is not intended to favor a particular merchant, and compatible services can participate without waiting for MoonPay to build a direct integration for every transaction. That matters for crypto because blockchain networks and stablecoins can provide programmable settlement without requiring every AI assistant, merchant, and financial provider to share one closed ecosystem. PayBox can also extend beyond conventional shopping into decentralized applications and other onchain services. MoonPay is not alone in pursuing this opportunity. Coinbase, Robinhood, Visa, and other financial companies are developing products that connect AI agents with payment or trading capabilities. PayBox’s potential advantage is its emphasis on open infrastructure and user-controlled funds rather than a payment experience limited to one marketplace. --- ### Why This Development Matters The real significance of PayBox is not that an AI can buy an item from Amazon or reserve a table. It is that AI assistants are beginning to receive the tools needed to take economic action instead of simply providing information. If this model gains adoption, the shopping process could shift from browsing individual websites to giving an assistant a goal, a budget, and a set of rules. Merchants would then compete to be discovered and selected by software agents as well as human shoppers. That future is not guaranteed. Agent-compatible merchants remain limited, consumers must learn to trust automated spending, and the industry still needs reliable standards for identity, fraud prevention, refunds, and accountability. PayBox nevertheless shows where the market is heading. Crypto payment rails may become most useful when they operate quietly underneath familiar AI interfaces, giving users programmable control without forcing them to understand every technical component. If MoonPay can make that experience secure and easy, PayBox could help turn agentic commerce from an industry concept into an everyday consumer tool.