ChatGPT Wallet lets agents check out; support and fraud stay with the seller

OpenAI launched ChatGPT Wallet on Friday 13 November 2026. A user saves a payment card once, and ChatGPT's agents, which OpenAI calls dots, can then choose a product and pay for it on the seller's own website, [TechCrunch reports](https://techcrunch.com/2026/11/13/chatgpts-new-wallet-lets-ai-agents-shop-and-check-out-for-you/). At launch it is open to a limited number of US ChatGPT Pro subscribers aged 18 or over with a US-issued Visa, Mastercard or American Express card.

For a company that sells online, nothing has to be switched on. OpenAI's [seller documentation](https://developers.openai.com/commerce/guides/accept-payments) says the Wallet "works out of the box with existing checkout pages and payment forms" with no integration, and that "your business handles the order, fulfillment, customer support, and fraud decisions." The agent arrives at your checkout whether or not you asked for it.

What reaches the support queue

Before the charge, the user reviews the items, the total and the card, and approves with Face ID or a fingerprint, which creates a record of the transaction, TechCrunch reports. The dot passes the seller the shipping address and contact details. Card rewards and the right to dispute a charge carry over.

On who answers when the agent orders the wrong thing, OpenAI's [agent access page](https://developers.openai.com/commerce/agent-access) says: "Customers are ultimately responsible for the actions they take, so any customer service or support should be handled in the same way you would if a customer took the action themselves." The same page says a site owner cannot yet require a person to confirm checkout on their own site; OpenAI says it requires human review "before any consequential action" for all users.

What reaches fraud and IT

• The seller does not get the card number. It gets a network token, a stand-in number issued by the card network, valid for that one purchase; OpenAI tells sellers not to save or reuse it.
• Bot filters may reject the sale. OpenAI says risk systems tuned to detect bots can be set off "by accident and cause purchases to fail". It is rolling out a way to pass the buyer's country, network and browser details to the seller in a web request header called Forwarded, and tells sellers to ask their fraud-screening provider whether it reads those signals.
• The agent's browser visits are signed, so a firewall can tell them from other bots. OpenAI says Akamai, Cloudflare, HUMAN and Vercel verify that signature automatically. An allow rule applies to the whole site, not to chosen pages or products.

What OpenAI has not said

The seller pages I read name no fee to the seller, no count of participating retailers, and no figure for failed purchases, disputes or returns. They do not say whether an order arrives marked as placed by an agent. TechCrunch reports that OpenAI plans spending limits and pre-approved purchases, which would remove the per-purchase review, with no date. Amazon blocks AI agents from its site, TechCrunch notes; United Airlines said on 11 November it would block them, and Target allows a named list, as [we reported](https://modelsatwork.news/article/united-airlines-blocks-customer-ai-agents-target-approved-list-walmart-agent-profiles).

What to do if you sell online

• Support: add a ticket tag for "customer says ChatGPT placed the order" this week, and write the refund rule for an agent's mistaken order before the first one arrives. OpenAI's position is that it is the customer's order.
• IT, or whoever runs the website's firewall: check whether your bot filter blocks the entry named ChatGPT Agent, and make that a decision, not an accident. Blocking loses the sale; allowing applies site-wide.
• Finance or fraud: ask your fraud-screening provider whether it reads OpenAI's Forwarded header signals, and how a declined or disputed Wallet order will show in your reports.