OpenAI launched ChatGPT Wallet on Friday, letting its AI agents — branded as dots — browse, compare and check out for Pro subscribers in the United States.
The rollout begins with a limited cohort of ChatGPT Pro users aged 18 and up who hold U.S.-issued Visa, Mastercard or American Express cards, TechCrunch reported.
Once a card is added, the dot can research products and stage a purchase, but the transaction stalls until the user reviews the merchant, itemized total and selected card, then authorizes it with Face ID or a fingerprint scan. That biometric step creates a signed record of consent for every order.
When the user approves, the payment network — not OpenAI — mints a dynamic token containing a virtual card number, expiry date and security code. The token travels to the merchant; the real primary account number never touches the dot or the retailer, Bangkok Post reported. American Express, Chase, Mastercard and Visa all back the scheme, so existing rewards, purchase protection and chargeback rights remain intact, CNBC-TV18 confirmed.
OpenAI stressed that the wallet does not replace the direct relationship between shopper and merchant. “These businesses retain control over pricing, order fulfilment, returns and customer service,” the company said in its launch blog. If a product is defective, the dispute flows through the card network exactly as it would for a manual checkout.
The largest retailer to ban AI agents from its site is Amazon, which wants traffic steered to its own Alexa assistant, TechCrunch noted. To smooth adoption elsewhere, OpenAI published a Commerce Partner Playbook urging merchants to expose structured product feeds, whitelist signed requests from known agents in their CDN and firewall rules, and coordinate with fraud-prevention vendors so legitimate agent-initiated transactions are not declined.
Today every purchase requires explicit biometric approval. OpenAI’s roadmap adds granular spending controls — category allow-lists, dollar caps and time-window permissions — so a dot could one day reorder household staples or book routine travel without waking the user.
No blanket liability wrapper covers agent errors. “OpenAI has not announced a blanket liability arrangement under which it would automatically compensate users for every problem arising from an agent-led purchase,” CNBC-TV18 wrote. Card-network protections apply, but gaps remain if a dot hallucinates a product spec or picks a non-refundable fare the user would have rejected.

Meta’s Muse has already logged over 10 million users with AI-assisted shopping as a top activity, while DoorDash and Uber report low agent-order volumes, TechCrunch found.
“We’re treating this limited launch as a live stress test for the entire agentic-commerce stack,” an OpenAI spokesperson told TechCrunch. The next expansion phase hinges on whether the Playbook convinces enough retailers to open their checkout APIs to non-human shoppers.