OpenAI launches ChatGPT Wallet: agents buy without ever seeing your card number
OpenAI has launched ChatGPT Wallet, letting an AI agent make purchases without ever seeing your card number. You add a card once, then before each purchase you review the items, the total and the card, and confirm with Face ID or a fingerprint. The merchant receives a token: a stand-in card number with its own security code and expiry date. The wallet currently works only for ChatGPT Pro subscribers in the US, so merchants should start preparing their sites for agents.
AI-processed from TNW; edited by Hamidun News
The key question about AI agents that can buy things is simple: who is responsible if the agent buys the wrong item, and if card data leaks? OpenAI answers it with the launch of ChatGPT Wallet, a wallet built into ChatGPT. The agent places the order, but neither it nor the merchant sees your card's real details. The service was built together with American Express, Mastercard and Visa.
How payment works
The scheme rests on two protections: human confirmation and a token instead of the card.
The user adds a card once. After that, before each purchase, ChatGPT shows the list of items, the total amount and the card the money will be charged to. The person confirms the order with biometrics, such as Face ID or a fingerprint. Each such confirmation is saved as a record of exactly what the user allowed the agent to buy.
After approval, the payment network issues a token. It is a stand-in card with its own number, security code and expiry date. The merchant and the agent work only with that token. To put it in everyday terms, it's like giving a courier a temporary entry pass to the building rather than the key to your apartment.
Cardholder terms stay the same. Bonuses and cashback remain, and disputes over things like a defective product are handled under the same protections as a regular payment. Prices, delivery, returns and support remain the merchant's responsibility.
Who can use it and what happens to the money
The launch is small for now. The wallet is available to ChatGPT Pro subscribers aged 18 and over in the US who have eligible cards from US banks. OpenAI says it is starting small to learn from real-world use. Chaya Nayak, who oversees product for monetization and payments at the company, described the goal as building a safe and simple foundation for merchants and retail partners. American Express's head of innovation, Luke Gebb, told CNBC that the limited rollout was a deliberate choice.
According to CNBC, OpenAI is not charging a commission on purchases for now. Looking ahead, the company expects users to hand purchases over to the agent entirely. To do that, a person would set limits by amount, category or time period. That mode does not exist yet: every purchase must be confirmed.
What stops agents from buying
The most down-to-earth part of the news concerns merchants. OpenAI acknowledges that sites may block agents by mistake. Anti-fraud systems can take a normal agent purchase for fraud, and then it gets declined or cancelled.
For businesses, the company has published a guide. It describes how to let ChatGPT agents through a firewall, upload product feeds and set up business profiles in ChatGPT. OpenAI is also working with anti-fraud companies Forter, Riskified, Sardine and Signifyd, and with payment providers Adyen, Checkout.com and Stripe.
Not everyone is ready to open the doors. United Airlines said it will ban agents from booking tickets. Amazon, according to CNBC, does not let them buy on its platform.
OpenAI has also had a bad experience before. In the fall of 2025 the company launched Instant Checkout, a payment feature built directly into ChatGPT, and fairly quickly backed away from it.
What it means in practice
For agent developers, the main lesson is: do not give the model real payment data. A token plus explicit human confirmation provides a clear trail of events and limits the damage when something goes wrong.
For online store owners, the takeaway is even more direct. Check whether your anti-fraud system is cutting off agent traffic. Prepare your product feed so the agent sees current prices and stock. It is also worth deciding in advance whether you will let agents in or block them, as United and Amazon have done.
The limitations are obvious. The service works only in the US, only for Pro subscribers and only with certain cards. Confirming every purchase does not amount to full automation, and spending limits remain a plan for now. How successful agents will be as buyers will become clear only with real-world use.
- Nov 21, 2026Anthropic mathematician says Claude helped prove the Riemann hypothesis in six weeks
- Nov 21, 2026Startup Osmo Trains AI on Smells Using 10 Million Human Ratings and Generates Fragrances From Text
- Nov 21, 2026North Korea Shows First Chatbot "Source of Wisdom": Science Only, No Politics
- Nov 21, 2026OpenAI Opens ChatGPT Spaces to Plus Subscribers: Shared Documents with AI
Key Questions Answered
How payment works?
The scheme rests on two protections: human confirmation and a token instead of the card. The user adds a card once. After that, before each purchase, ChatGPT shows the list of items, the total amount and the card the money will be charged to. The person confirms the order with biometrics, such as Face ID or a fingerprint. Each such confirmation is saved as a record of exactly what the user allowed the agent to buy. After approval, the payment network issues a token.
Who can use it and what happens to the money?
The launch is small for now. The wallet is available to ChatGPT Pro subscribers aged 18 and over in the US who have eligible cards from US banks. OpenAI says it is starting small to learn from real-world use. Chaya Nayak, who oversees product for monetization and payments at the company, described the goal as building a safe and simple foundation for merchants and retail partners. American Express's head of innovation, Luke Gebb, told CNBC that the limited rollout was a deliberate choice.
What stops agents from buying?
The most down-to-earth part of the news concerns merchants. OpenAI acknowledges that sites may block agents by mistake. Anti-fraud systems can take a normal agent purchase for fraud, and then it gets declined or cancelled. For businesses, the company has published a guide. It describes how to let ChatGPT agents through a firewall, upload product feeds and set up business profiles in ChatGPT.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.