If staff or customers start asking AI agents to buy things, your payment policy, fraud rules and vendor reviews will need an answer for a new kind of buyer. OpenAI’s ChatGPT Wallet is an early, limited version of that shift. It is small today, but it sets the pattern: the agent shops, the human approves, and a token stands in for the card.
What happened
OpenAI launched ChatGPT Wallet, which lets its Dots agents make purchases from a card the user has added. According to The Verge, the agent checks in before each checkout, and OpenAI’s support page says every purchase needs a biometric security check such as Face ID or a fingerprint scan.
The Verge also reports that once the user approves, OpenAI swaps the real card details for a payment token, much like tap to pay, Apple Pay or Google Wallet. Neither the agent nor the merchant sees the actual card number. 9to5Mac says the launch includes a partnership with American Express, Mastercard and Visa, and TechCrunch says the rollout starts with U.S. Pro subscribers only.
OpenAI says it expects users eventually to delegate purchases fully, setting limits such as spending caps, categories or time periods. The Verge notes that some retailers restrict or block AI agents, and that Amazon blocked Meta’s Muse agents earlier this year.
Why it matters
The design borrows trust from systems that merchants and card networks already know: tokenization plus a biometric confirmation. That suggests agent commerce will grow through existing payment rails rather than new ones, with the card networks involved from the start.
The harder problem is acceptance. OpenAI itself says websites must avoid blocking agents and that payment and risk systems must recognise legitimate agent purchases. Fraud tools built to flag automated traffic may treat a valid agent checkout as an attack. Retailers face a real choice between opening up and keeping control of the customer relationship, as we saw when United blocked AI agents from completing bookings.
The planned move to pre-approved, rule-based buying is the bigger shift. Once a human no longer approves each purchase, spending limits and category rules become the main control, and they start to look like corporate card policy.
What it means for you
Update your payment policy before staff improvise. Decide whether employees may use personal agent wallets for work purchases, and whether company cards may be added to one. Today’s per-purchase biometric check keeps a person in the loop, so treat that as the baseline you require.
Ask vendors direct questions. If you sell online, check whether your bot protection and fraud scoring would block a legitimate agent checkout, and who owns that decision. If you buy software that touches payments or procurement, ask how it handles agent-initiated transactions, tokens, audit logs and dispute handling.
Plan for accountability. When an agent picks the wrong item or vendor, someone must own the error. Define who reviews receipts, how refunds work and what evidence you keep of the approval step.
Do not budget for scale yet. A limited U.S. Pro rollout means this is a pilot topic, not a rollout project. Track pricing changes for agent features on AI Tool Radar and follow our AI coverage for wider availability.
What to watch
- Whether OpenAI ships the delegated mode with spending limits, categories and time windows, and how granular the controls are.
- Which large retailers allow agents and which follow Amazon’s earlier block of Meta’s agents.
- How Visa, Mastercard and American Express update risk rules and dispute handling for agent purchases.
- Expansion beyond U.S. Pro subscribers, especially to business plans.
