1. OpenAI launches the ChatGPT Wallet
OpenAI on Friday, November 13, launched the ChatGPT Wallet, a feature that allows AI agents inside ChatGPT to complete purchases on a user's behalf through existing card networks. The company built the wallet together with American Express, Mastercard and Visa, and is rolling it out first to ChatGPT Pro subscribers in the United States with eligible U.S.-issued cards.
The launch marks a step beyond the role AI assistants have mostly played in online shopping so far. Instead of recommending products and sending users off to a merchant's website to finish the transaction, ChatGPT's agents can now carry the purchase through to payment, with the user approving each transaction along the way.
OpenAI has also outlined where it wants the feature to go next. The company plans to eventually let users authorize purchases within preset spending limits, so that agents could complete transactions without asking for approval each time.
2. What is the ChatGPT Wallet?
The ChatGPT Wallet is a payment tool inside ChatGPT that stores a user's eligible card and lets the chatbot's AI agents use it to buy goods and services, with each purchase confirmed by the user through biometric verification. In practice, a user asks an agent to find and buy something, and the agent handles the steps up to checkout before asking for approval.
That approval is the critical safeguard. Users confirm transactions with biometric checks such as a face scan or a fingerprint, rather than by typing a password or card number. Only after that confirmation does the purchase proceed.
The wallet relies on the infrastructure that already moves card payments around the world. Rather than creating a new payment system, OpenAI has connected its agents to the networks operated by its three card partners, which already process transactions for millions of merchants.
3. How the payments are secured
Card details are never shared with the agent or the merchant. For every approved purchase, the card networks generate a tokenized credential, a stand-in for the real card number that is tied to that specific transaction. The merchant receives the token, completes the sale and gets paid, but never sees the underlying card information, and neither does the AI agent.
Tokenization is a well-established technique in payments. It is the same basic approach that powers mobile wallets on smartphones, where a device-specific token replaces the card number at the point of sale. Its value lies in limiting the damage if a merchant's systems are breached: a stolen token tied to one transaction is far less useful to criminals than a full card number.
Applying the method to AI agents addresses one of the most obvious worries about letting software spend money. If an agent never holds a reusable card number, the risk that a compromised or misbehaving agent could leak payment details is sharply reduced. Combined with biometric approval for each purchase, the design keeps the user firmly in control of when money leaves their account.
4. Who can use the ChatGPT Wallet?
At launch, the wallet is limited to ChatGPT Pro subscribers in the United States, and only cards issued in the country that meet eligibility requirements can be added. Pro is the most expensive tier of ChatGPT's consumer subscriptions, which means the first users are likely to be the platform's most engaged customers.
Starting with a narrow group is a common approach for payment products, which must satisfy security, fraud and compliance requirements before they can scale. A limited release lets OpenAI and its card partners observe how agents behave in real transactions, how often users approve or reject purchases, and where problems arise, before opening the feature to a wider audience.
5. Plugging into existing checkout systems
OpenAI says it is working with payment processors and fraud-prevention companies so that purchases made by AI agents can flow through the checkout systems merchants already use. That point matters for adoption. Retailers are unlikely to rebuild their payment setups for a new category of buyer, so making agent-initiated purchases look and behave like ordinary card transactions lowers the barrier for merchants to accept them.
Fraud prevention is a particular concern. Card networks and processors use detailed signals to decide whether a transaction is legitimate, and a purchase initiated by software rather than a person sitting at a keyboard could trigger those systems in unexpected ways. Working with fraud-prevention firms is intended to help those systems distinguish an authorized agent acting for a verified user from a bot attempting to commit fraud.
The approach also benefits the card partners and processors involved. Every agent-initiated purchase that runs through their systems is a transaction they can process, secure and monitor with tools they already operate, rather than volume that might otherwise migrate to new payment methods built specifically for software agents.
6. What the launch means for shoppers
For consumers, the main promise is convenience. Many online purchases involve repetitive steps: searching across sites, comparing prices and options, creating accounts, and entering shipping and payment details. An agent that handles those steps and returns with a ready-to-approve purchase could save time, especially for routine or well-defined orders.
The design also tries to address the trust problem that has held back automated buying. People are understandably cautious about giving software the ability to spend money. By requiring biometric approval for every purchase at launch and keeping the real card number hidden from both the agent and the merchant, OpenAI is betting that users will be willing to try the feature because they retain the final say over each transaction.
Because purchases run over existing card networks, shoppers are dealing with the same kind of card payment they already use online, rather than a new and unfamiliar payment method. That familiarity could make the transition easier for users who are curious about AI agents but wary of untested financial tools.
There are open questions as well. Users will need to judge how well an agent understands their preferences, how it chooses between competing products and merchants, and how mistakes are corrected. Those questions will become more pressing once purchases no longer require individual approval.
7. From recommendations to completed transactions
The wallet reflects a broader shift among AI companies, which are moving from suggesting products to actually completing purchases. This emerging category is often called agentic commerce: AI agents that act on a user's instructions to search, compare and buy.
If the model gains traction, it could change how consumers interact with online merchants. A shopper who delegates a purchase to an agent may never visit the retailer's website, see its promotions or browse related items. That would shift some influence over buying decisions away from merchants and toward the AI platforms that their customers use to shop.
For OpenAI, the feature deepens ChatGPT's role in users' daily lives. A chatbot that can research a purchase and then pay for it becomes a more complete assistant, and payments give the company a direct role in commercial activity rather than only in conversations about it.
8. Card networks and tech rivals chase the same market
OpenAI is entering a field where several large players are already active. Google released its Agent Payments Protocol in 2025, a framework intended to let AI agents carry out transactions on behalf of users. Visa and Mastercard have each launched their own initiatives aimed at supporting payments initiated by AI agents.
The participation of Visa and Mastercard as partners in the ChatGPT Wallet shows how the card networks are positioning themselves. Rather than being bypassed by new payment methods built for AI, they are seeking to become the default rails for agent-initiated spending, extending their tokenization and fraud tools into a new channel. American Express, the third partner, brings its own network and cardholder base to the arrangement.
9. Crypto's alternative vision for AI agent payments
A parallel effort is unfolding in the crypto industry, where developers argue that blockchain-based payments are better suited to software agents. In 2025, Coinbase introduced its x402 protocol, which allows autonomous agents to pay for digital services with stablecoins.
The appeal of stablecoins for this use case lies in their programmability. Payments can be sent directly between software programs at any hour, in small amounts, and with rules embedded in code, which supporters say fits the way autonomous agents operate better than card payments designed for human shoppers.
OpenAI's choice to build on traditional card networks highlights the competition between the two approaches. Card rails offer immediate reach to merchants and familiar consumer protections, while crypto-based systems promise flexibility and open access. The ChatGPT Wallet puts the weight of one of the most widely used AI products behind the card-based model, at least for its first step into payments.
10. What comes next for the ChatGPT Wallet
The most significant change on OpenAI's road map is the move to preauthorized spending. Allowing agents to buy within limits set in advance, without approval for each purchase, would make them far more autonomous, but it would also raise new questions about how users set boundaries and how disputes are handled when an agent buys the wrong item.
In the near term, the key questions are how quickly OpenAI expands the wallet beyond Pro subscribers and the United States, and how merchants and fraud systems respond to a growing volume of agent-initiated purchases. The answers will help determine whether agentic commerce becomes a routine way to shop or remains a niche feature for early adopters.