- OpenAI says ChatGPT Wallet lets agents pay with a token instead of real card details, after the user approves each purchase with a security check.
- OpenAI warns that sites and risk systems may block or decline legitimate agent purchases. Its playbook asks businesses to review bot access and fraud settings.
- Leaders should ask who owns agent access to checkout, and how their payment and risk providers treat agent purchases.
The shopper at your checkout may not be the one who clicks
For years, a checkout page has assumed a person on the other end. Fraud tools have largely been built around that assumption, so unusual behaviour looked like risk. OpenAI's ChatGPT Wallet, announced on November 13, challenges it. A software agent can now pay on a customer's behalf, and the business has to decide whether that agent is welcome.
What OpenAI announced
OpenAI says it built ChatGPT Wallet with American Express, Mastercard and Visa. The first group of users is narrow. Access begins inside OpenAI's dots product and covers ChatGPT Pro subscribers who are at least 18, live in the US and hold a qualifying card issued there.
OpenAI says 1.2 billion people use ChatGPT each week, and that many use it to research and compare products. Wallet is meant to help them finish the purchase. It also calls this a limited start, so it can learn from real use.
How the payment works
The user adds a card once. For every purchase, the agent shows the user what is in the basket, what it costs and which card will pay. The user confirms with biometrics, for example Face ID or a fingerprint scan.
After approval, the payment network issues a payment token. OpenAI describes it as a substitute for the real card details, with its own card number, security code and expiration date. OpenAI says the business and the agent do not receive the actual card details.
Each approval also creates a record of what the user authorized. OpenAI says the card keeps its rewards and its protections for eligible disputes, such as a defective item.
Think of a valet key. It lets someone do one job without handing over everything. The design separates the power to buy from the power to hold the secret.
Where the work lands on merchants
OpenAI says the tokens work with existing checkout flows. It adds that risk systems may still flag legitimate agent purchases, leading to declines or cancellations. It also says some sites may unintentionally block purchases made through Wallet.
Its Commerce Partner Playbook lists four areas for businesses. First, claim your business in ChatGPT and upload a product feed with current prices and availability. Second, set your CDN or firewall to recognise signed requests using Web Bot Auth. Third, talk to your payment and risk providers about agent purchases. Fourth, optionally build a plugin or add WebMCP tools so agents can check stock or manage carts.
Signing lets a bot prove who sent a request. The playbook asks the merchant's CDN or firewall to recognise these signed requests, so the traffic is identified as coming from a ChatGPT agent. It also asks businesses to confirm that their general security settings are not blocking those agents by accident.
OpenAI says it is working with fraud-prevention firms Forter, Riskified, Sardine and Signifyd. It names Adyen, Checkout.com and Stripe among the payment providers. The aim is for their systems to recognise legitimate agent purchases.
The idea: trust moves from the person to the permission
When the buyer is software, "is this a human?" stops being the useful question. The useful question is "did a human authorize this, and can we prove it?" Wallet's answer is a per-purchase approval and a record of it.
That shifts a burden. Until now, bot defences were built to keep automated traffic out. A store that wants agent sales now needs rules that let the right agents in. The risk is a quiet one. A sale lost to a blocked agent or a fraud-system decline may never show up as a complaint.
Several limits apply. This is OpenAI's account of its own product, and the announcement gives no decline rates or independent testing. It is one product launch, so it shows what one company is asking of merchants, not a market trend. OpenAI also says it expects users to eventually delegate purchases within set limits, such as spending caps or categories. That is a stated plan, not a feature in this release.
Questions to put to your team
Ask who owns agent access to checkout. In many firms, security teams tune firewalls to block bots, and e-commerce teams are paid on conversion. The two goals now meet in one setting.
Ask your CDN or firewall provider whether it can recognise signed requests using Web Bot Auth. Ask your payment and risk providers how they tell a legitimate agent purchase from fraud. Ask whether your product feed shows current prices and availability.
Then agree how you will spot declines and cancellations that involve agents, so the cost is visible. A checkout built to keep bots out now has to learn which ones to let in.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: OpenAI.
