OpenAI has introduced ChatGPT Wallet, a US-first payment feature that lets ChatGPT agents make purchases without seeing a user's real card details. Each purchase still needs the user’s review and a security check, putting consent between an AI recommendation and the checkout button.
OpenAI Watch analysis
What happened
ChatGPT Pro subscribers in the US can use the wallet through ChatGPT’s dots feature, according to GHacks. OpenAI built it with American Express, Mastercard and Visa.
For each purchase, the user reviews the items, total and selected card, then approves with a security check such as Face ID or a fingerprint scan. The payment network creates a token, a substitute card number, so the business and agent do not receive the real card details. Each approval also creates a record of what the user authorised.
OpenAI says the wallet is starting with limited availability while it works on compatibility with businesses, banks and payment networks. It has not said when the feature will reach Plus or free subscribers, or users outside the US.
Why it matters
This is a step from AI helping people choose what to buy to helping complete the purchase. The current design keeps a person in the loop for every transaction, while tokenisation separates the agent and retailer from the underlying card details. Those are meaningful guardrails, not a guarantee that every dispute, mistaken order or security problem disappears.
OpenAI says users retain card rewards and protections for eligible disputes, while businesses remain responsible for pricing, delivery, returns and customer service. The arrangement makes the agent a new route to checkout, rather than the shop itself.
Our read
The useful distinction is between an agent that prepares a purchase and one that can spend unattended. Today’s version asks for approval each time; OpenAI says it plans to offer advance-approved purchases within user-set limits for spending, categories and time periods. That future version will need to earn trust one boundary at a time. A shiny new checkout button is easy. Making it clear what the agent may buy, and what happens when it gets things wrong, is the harder work.
What to watch
- Whether OpenAI publishes a timetable for wider access beyond US ChatGPT Pro subscribers.
- How spending limits and advance approvals work if OpenAI rolls them out.
- Whether merchants and payment providers adopt the system, and how disputes are handled.
Discussion spark: Would you let an AI agent buy within limits you set in advance, or should every purchase require a fresh approval?
Sources and evidence
- Source update (15 November 2026, 02:36 UTC)
- Source update (15 November 2026, 02:36 UTC)
- Source update (15 November 2026, 02:36 UTC)
OpenAI Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by OpenAI.