ChatGPT Wallet lets OpenAI's dots agents buy things for you, with your fingerprint

OpenAI's dots agents can now pay for things. The company on Friday began rolling out ChatGPT Wallet, a payment feature that lets the agents complete checkout using a stored card. Before any money moves, the user has to approve the purchase with a biometric check.

The first wave is narrow. OpenAI is offering the wallet to a limited number of ChatGPT Pro subscribers in the US who are 18 or older, have a US billing address and hold an eligible US-issued Visa, Mastercard or American Express card. The company built the feature in partnership with those three card networks.

OpenAI launched dots, its platform for personal AI agents, earlier this year. The wallet is the piece that lets those agents finish a purchase through ChatGPT instead of stopping before checkout.

How a purchase works

Setup happens once. A subscriber adds an eligible card to ChatGPT Wallet and doesn't have to enter it again. From then on, the user can ask a dot to help buy something, and the agent handles the shopping up to the point of payment.

At checkout the agent stops. The user sees the items, the total and the card that will be charged, then signs off with Face ID, a fingerprint scan or a similar security check. OpenAI's support documentation says every purchase made through the wallet requires that check, and each approval produces a record of what the user authorized the agent to buy.

The payment itself relies on tokenization, the same technique behind tap-to-pay, Apple Pay and Google Wallet. Once the user approves, the card network issues a stand-in card number, security code and expiration date, and the charge goes through on that token. The real card details are never shared with the dot or with the merchant. The merchant gets what it needs to fill the order, including the shipping address and contact information, from the agent.

Purchases made this way still earn card rewards, and OpenAI says they keep the usual card protections for eligible disputes. That matters more than it might seem. If an agent buys the wrong size or the wrong product, the ability to dispute the charge through the card issuer is the safety net most shoppers already count on.

OpenAI is marketing the feature hard on security. The company says the wallet is "setting a new bar for secure payments by agents," and calls it the first wallet that lets agents pay without access to a user's actual card details while also requiring a security check on every purchase. That "first" is OpenAI's own claim. The design choices behind it hold up, though. Tokenization is proven technology, and per-purchase biometrics close off the most obvious failure: an agent spending money nobody approved.

The tradeoff is that this version isn't very autonomous. If you have to review the cart and scan your face before every order, the agent saves you browsing and form-filling, not decision-making. For a first release, that's a reasonable place to draw the line. It is also a modest step for a product pitched as an agent.

Where the agents can shop

The wallet's bigger limit has nothing to do with payments. It only works on sites that let agents in, and a growing number don't.

Amazon blocked Meta's Muse agent from shopping on its site, according to a GeekWire report last month. This week United Airlines announced a policy restricting agent access to its digital platforms, both directly and through travel agencies and aggregators. OpenAI acknowledges the problem. The company says that for agents to use the wallet, websites have to make sure agents aren't inadvertently blocked, and payment and risk systems have to recognize legitimate purchases made by agents.

To push things along, OpenAI published a Commerce Partner Playbook alongside the wallet. The company describes it as tools and practical guidance for businesses that want to support customers using agents. A guide can't persuade a retailer whose business depends on owning the customer relationship, though. Amazon's block of Muse shows how quickly a major store can shut the door, and if the largest shopping site stays closed, many everyday purchases will remain out of an agent's reach no matter how secure the wallet is.

OpenAI also frames the Pro-only launch as a way to work on these issues with partners. "We're starting with limited availability to ChatGPT Pro users," the company said, adding that it wants to "learn from real-world use and work with businesses, banks, and payment networks." Starting with Pro, ChatGPT's top-tier plan, keeps the first users to a small, committed group. That limits the damage if something breaks, and it means early feedback will come from heavy users rather than typical shoppers.

The plan for hands-off buying

OpenAI is clear that the confirm-every-purchase model is a starting point. The company said it expects users eventually to "delegate purchases entirely to their agent by setting clear parameters," such as spending limits, purchase categories or time periods, and to pre-approve those purchases. It said that future setup would come with stronger safeguards built on today's protections, including biometric checks and records of authorization.

That is the version that would actually change how people shop: an agent allowed to reorder household basics under a monthly cap, or to book a flight when the fare drops below a set price. It's also where the hard questions about trust begin. OpenAI hasn't said when pre-approval will arrive, and nothing in its announcement explains how disputes would work for a purchase no person reviewed at the moment it happened.

Consumer demand for any of this is still uneven. DoorDash and Uber have reported low numbers of orders placed by AI agents, while Meta has seen AI-assisted shopping become a top activity on Muse, which now has more than 10 million users. Other personal agents, such as Instinct, already support buying on a user's behalf. OpenAI is arriving in a market where shopping looks like one of the few agent tasks people use regularly, and where rivals have a head start.

Shopper confidence is the other open question. Many people have read about AI models getting facts wrong, and handing one a payment method asks for far more trust than asking it for a recipe. Mandatory biometric approval and preserved dispute rights look designed to win over cautious users first. Whether OpenAI can relax those controls later without losing that trust is the real test for ChatGPT Wallet, and the company hasn't set a date for it.