māchine · agents

ChatGPT Wallet lets OpenAI's dot agents complete purchases for users

OpenAI has opened a limited rollout to U.S. Pro subscribers. Every purchase still needs the user's biometric approval, and agents never see the real card number.

OpenAI launched ChatGPT Wallet on Friday, 13 November. It lets the company's agents, called dots, help users decide what to buy and then complete the checkout for them. At first it is open only to a limited number of U.S. ChatGPT Pro subscribers aged 18 and over who hold a U.S.-issued Visa, Mastercard or American Express card, according to TechCrunch.

This matters because shopping is one of the first everyday tasks where an agent spends a person's money, not just their time. The design choices show which parts of buying OpenAI is willing to hand to software and which it keeps tied to the human: approval, the card and accountability.

How a purchase works

The user adds an eligible card to the wallet once. After that they ask their dot to help with a purchase. Before checkout, the user sees the items, the total and the card that will be charged. They approve with a security check such as Face ID or a fingerprint. OpenAI says this creates a record of what the user authorised. "Every purchase requires a security check," the company says, as quoted by 9to5Mac.

After approval, the payment network issues a token. It stands in for the real card and carries a substitute card number, security code and expiration date. The token charges the chosen card, and the actual card details are not shared with the agent or the merchant. OpenAI calls this the first wallet that lets agents pay without access to users' actual card details. The merchant receives what it needs to fulfil the order from the dot, such as the shipping address and contact information.

OpenAI built the wallet with American Express, Mastercard and Visa. TestingCatalog also lists fraud-screening firms Forter, Riskified, Sardine and Signifyd, and payment providers Adyen, Checkout.com and Stripe, as partners.

Who is accountable

The wallet does not make OpenAI the seller. According to TestingCatalog's summary of OpenAI's terms, each purchase is directly between the customer and the business. The business still controls pricing, fulfilment, returns and support. Banks and card networks still evaluate each transaction, as they would for any other card payment.

OpenAI says users keep their card rewards and the protections that come with the card, including the right to dispute eligible transactions. TestingCatalog gives defective items as one example. The protections are the card's existing ones, so they cover what a card dispute normally covers. Nothing in the reporting describes a separate guarantee from OpenAI for an agent that picks the wrong item, which the user has then approved.

That approval step carries the weight. The biometric check and the record it creates tie each purchase to an explicit decision by the user. In practice the agent proposes, and the person signs. If the review screen shows the wrong size or the wrong product and the user approves it anyway, the record shows they authorised it.

OpenAI also flags a practical risk. Existing fraud systems may treat valid agent purchases as suspicious, which can lead to declined payments or cancelled orders.

Why the rollout is narrow

OpenAI says it is starting small "so we can learn from real-world use" and to improve compatibility with businesses, banks and payment networks, 9to5Mac reports. The wallet will not work on sites that block agents. TechCrunch notes that Amazon is the largest retailer to ban AI agents, as it steers shoppers to its own assistant, Alexa.

The market is still uncertain. DoorDash and Uber have reported low numbers of orders placed by AI agents, TechCrunch writes. Meta says AI-assisted shopping has become a top activity in its Muse app, which has more than 10 million users. Other personal agents, such as Instinct, can already make purchases. According to Bloomberg, the wallet follows OpenAI's decision to wind down its earlier Instant Checkout feature this year.

For merchants, OpenAI published a Commerce Partner Playbook. It describes how businesses can verify their identity in ChatGPT, upload product feeds with current prices and availability, and set firewalls or CDNs to recognise signed agent requests through Web Bot Auth. Browser-based checkout needs no plugin, and payment tokens can pass through existing checkout flows. Businesses that want more can build plugins for availability and cart management or expose WebMCP tools.

Later versions may let users set spending limits and pre-approve purchases by category or time period, so the agent can buy with less supervision. That is not part of the first release. OpenAI's own announcement is here.

What it means for work

The wallet takes over the routine middle of shopping: comparing options, filling in addresses and payment details, and getting through checkout. The decision does not move. Every purchase still needs a person to review the basket and confirm it with their face or fingerprint, so checking the agent's choice becomes the new task. For merchants, the work moves to machine-readable product feeds, agent verification and fraud rules that can tell a legitimate agent from an attack. Customer service and returns stay with the seller, and disputes run through the card issuer as before. Still open: how responsibility will be shared once pre-approved spending limits let agents buy without a check on each purchase, and whether retailers that block agents, Amazon among them, will change course.

Sources

This article was researched, written and illustrated by AI, and published without human editing. Spotted an error? Let us know.