ChatGPT Wallet lets OpenAI's AI agents buy things for you, with your fingerprint as the last step

The last step before an AI agent spends your money is now a fingerprint. On 13 November, OpenAI announced ChatGPT Wallet, a way for agents in ChatGPT Work, Codex and its “dots” assistants to complete purchases once the user approves them.

Shopping by AI agents, sometimes called agentic commerce, is where software meets real money, and where trust is hardest to win. Dots, which OpenAI introduced on 29 September, are proactive assistants that keep working on projects and everyday tasks. Paying is the step that turns them from helpers into buyers, which is why the safeguards matter more than the feature.

How ChatGPT Wallet asks before it spends

ChatGPT Wallet starts small. TechCrunch and The Next Web report that it is open first to a limited number of ChatGPT Pro subscribers in the US, aged 18 or older, with an eligible US-issued Visa, Mastercard or American Express card. OpenAI is starting small to learn before a wider rollout.

The flow is simple. You add a card once and ask your agent to help you buy something. Before checkout, you review the items, the total and the card, then approve with Face ID or a fingerprint. Each approval creates a record of what you allowed.

OpenAI says the card network issues a token, a stand-in card number, so neither the agent nor the merchant sees your real card details. TechCrunch adds that the agent passes the merchant what it needs, such as a shipping address and contact details. Card rewards and standard protections, such as disputes for eligible issues, still apply, according to TechCrunch and The Next Web. Merchants keep control of pricing, delivery, returns and customer service.

The shops still have to say yes

Payment is only half the puzzle. Some retailers block AI agents, and Amazon is the largest to ban them, TechCrunch and CNBC reported. United Airlines has said it will block agents from booking, as we covered in our story on United’s ban on AI agents booking flights.

The early numbers are cautious. TechCrunch reports that DoorDash and Uber have seen low numbers of agent orders. The Next Web says Visa tested 22 AI models as shoppers this month and they failed 42.8% of the time. We have not seen Visa’s own account of that test, so treat the figure as reported.

OpenAI has been here before. CNBC reported that Instant Checkout, launched in autumn 2025, was later set aside. This time the company says it is focused on merchants and network partners. “Our goal here is to really focus on merchants and our network partners,” said Chaya Nayak, OpenAI’s product lead for monetization and payments, as quoted by The Next Web.

OpenAI also published a playbook for businesses on product feeds, business profiles and letting agents through firewalls. The Next Web names fraud firms Forter, Riskified, Sardine and Signifyd, and payment providers Adyen, Checkout.com and Stripe, as partners.

CNBC reported that OpenAI is not taking a cut of purchases for now.

If you are outside the US, or not on Pro

For most readers, nothing changes today. The launch is US-only and limited to some Pro subscribers, and OpenAI has not given a date for a wider rollout, TechCrunch notes. If you do get access, the sources suggest a habit worth keeping: read the total and the card on the approval screen every time, not just the product.

OpenAI says it plans spending limits and rules by category or time period. It also plans to let users pre-approve some purchases so agents can act with more autonomy. These are plans, not features you can use now. If your own business sells online, the playbook is where to look.

Why a fingerprint per purchase is the right place to start

There is real good here. A person approves every purchase. The agent never holds your real card number. Existing card protections still apply. Those three choices answer the questions most shoppers would ask first.

The weak points are just as clear. Pre-approved purchases, when they come, would loosen the human check, so the limits will matter a great deal. Retailers can still say no. And an agent that fails nearly half the time in a test, if that figure holds up, is not ready to run your errands unwatched.

On balance, OpenAI has chosen caution, and that is a sensible way to start. Luke Gebb of American Express told The Next Web it makes sense to start in a contained environment and keep rolling from there. The likelier outcome is steady, careful growth, with trust built one approved purchase at a time.

The next things to watch are the spending limits, and whether the big shops open their doors to agents.