ChatGPT Wallet Lets Agents Check Out: Who Can Use It Now
OpenAI's Wallet lets dots agents complete purchases with Face ID approval and tokenized cards. Limited to U.S. Pro users, and Amazon blocks agents.

OpenAI shipped ChatGPT Wallet on Friday, letting its agents buy things on your behalf — starting with a limited group of U.S. Pro subscribers. That closes a practical gap in agentic shopping: an agent that recommends a product can now also complete the transaction, which OpenAI frames as the point of the feature.
Who can actually use it today
The rollout is narrow. Per TechCrunch's report, the option is "initially be available to a limited number of U.S. ChatGPT Pro subscribers, ages 18 and up, with a U.S.-issued Visa, Mastercard, or American Express." Three conditions stack there — U.S. residency, a Pro subscription, and a card from one of three networks — so most readers checking their app this weekend will not see it.
The agents doing the buying are what OpenAI calls dots. OpenAI's own announcement describes Wallet as a way for agents "in ChatGPT Work, Codex, and dots to complete purchases with user approval and existing card protections," which puts the same payment rail behind its developer and workplace surfaces, not just consumer chat.
Setup is one-time. Users add eligible cards to the Wallet, then ask a dot for help with a purchase; TechCrunch notes they do not have to re-enter payment information on every buy.
How the money actually moves
The mechanics matter more than the demo here, because they decide what you are exposed to when the agent gets it wrong.
Three things happen at checkout, according to TechCrunch's description of the feature:
- You approve before the charge. Users can review the items, the total, and the card before checkout occurs, and approve "directly with a security check like Face ID or a fingerprint, which creates a record of their transaction."
- The card number never reaches the agent or the store. Wallet charges the card using a payment token, described as "a substitute for the user's actual card details, so that information isn't shared with either the dot or the merchant."
- The merchant still gets fulfilment data. Shipping address and contact information pass through automatically from the dot.
Two consumer protections survive the hop, which is the detail most likely to decide adoption: TechCrunch reports that customers shopping via dots still earn credit card rewards and keep the usual card protections, including the ability to dispute transactions. An agentic purchase that forfeited chargeback rights would be a hard sell; this one does not.
What is missing at launch
Spending controls are not in the first version. OpenAI "plans to offer users more control over the feature by allowing them to set spending limits or configure other parameters around purchase categories or time periods," per TechCrunch — the goal being pre-approved purchases the agent can then handle more autonomously. Until that ships, per-purchase biometric approval is the only brake, which caps how unattended any of this can be.
Retailer coverage is the other open question. OpenAI says it aims to support a broader number of retailers over time.
The retailer problem is the real constraint
Here is where the strategy gets interesting. TechCrunch points out that some retailers are blocking AI agents from using their sites outright, and names the biggest one: "on the shopping front, the largest retailer to ban AI agents is Amazon, as it aims to steer users to its own AI agent, Alexa, instead."
That is a structural problem, not a bug OpenAI can patch. The company's answer is a Commerce Partner Playbook, published alongside the launch, offering tools and guidance to sites that want to support customers shopping through agents. Read that as what it is: a bid to make agent-readable checkout a thing merchants opt into, in the same way they once opted into mobile-friendly pages.
Is there demand for this yet?
The evidence is genuinely split, and TechCrunch cites both sides. Delivery platforms are not seeing it: DoorDash and Uber "have so far reported low numbers of orders placed by AI agents." Social is seeing something different — Meta's Muse, with over 10 million users, has seen AI-assisted shopping become a top activity on the app. Other personal agents, such as Instinct, also support purchases.
The gap between those two signals is worth sitting with before concluding anything. Buying a burrito through an agent saves nobody any time; the ordering apps already solved that. Being talked through a purchase you were unsure about is a different job, and it is the one the Meta number appears to reflect. OpenAI's framing — the agent helps you decide what to buy, then completes it — lands closer to the second.
What to watch next
Three markers will tell you whether this becomes infrastructure or a Pro-tier curiosity. First, whether spending limits and category rules ship fast enough to allow genuinely unattended purchases. Second, how many merchants adopt the Commerce Partner Playbook versus how many follow Amazon's lead and block agents. Third, whether eligibility widens past U.S. Pro subscribers with cards from three networks — an agentic checkout that only works for the top tier in one country is a pilot, not a payment rail.
If you run a storefront, the useful move this week is not to pick a side. It is to check your own logs for agent traffic and read the playbook before deciding whether to court it or block it.
More from DangMua