OpenAI has introduced ChatGPT Wallet, a payment feature that allows AI agents to make purchases on behalf of users without accessing their actual card details.
The feature requires users to approve every transaction through a security check, such as Face ID or fingerprint authentication.
Built in partnership with American Express, Mastercard, and Visa, ChatGPT Wallet will initially be available to ChatGPT Pro subscribers aged 18 and above in the US with eligible US-issued cards. The rollout will begin through OpenAI’s Dots personal agent.
The move expands ChatGPT’s role in online shopping, allowing users to move from researching and comparing products to completing purchases through AI agents.
With ChatGPT Wallet, users add a payment card once and authorise individual purchases without repeatedly entering their payment details. Before completing a transaction, the agent presents the items, total cost, and selected payment card for the user to review and approve.
Following approval, the payment network generates a payment token containing a substitute card number, security code, and expiration date. This allows the transaction to be processed without sharing the user’s actual card details with the agent or merchant.
OpenAI said the system also preserves existing card rewards and protections for eligible disputes with businesses, including cases involving defective products.
Users will continue to transact directly with merchants, which retain control over pricing, fulfilment, returns, and customer service.
“ChatGPT Wallet marks our first American Express Agentic Commerce Experiences integration, enabling eligible Card Members to use their Amex Card while reviewing and approving each purchase,” said Luke Gebb, EVP and Head of Global Innovation at American Express.
Alongside the wallet, OpenAI is releasing a Commerce Partner Playbook to help businesses adapt their websites, payment systems, and risk management processes to purchases made through AI agents.
The company said some websites may inadvertently block agent-led purchases because their security systems are not configured to recognise legitimate activity from AI agents.
The playbook guides on making product information accessible, allowing agents to access websites, and working with payment providers to reduce transaction declines.
Businesses can upload product feeds containing current pricing and availability after claiming and verifying their business on ChatGPT. They can also configure content delivery networks and firewalls to recognise signed requests through Web Bot Auth.
OpenAI is also working with fraud prevention companies, including Forter, Riskified, Sardine, and Signifyd, as well as payment service providers such as Adyen, Checkout.com, and Stripe, to help identify legitimate agent-led transactions.
Businesses can also build plugins that allow agents to check product availability, manage shopping carts, and interact with their existing systems. OpenAI has also highlighted WebMCP tools as another way for websites to support agent interactions.
OpenAI said it eventually expects users to delegate purchases entirely to AI agents by setting parameters such as spending limits, product categories, and time periods.
Such transactions could be pre-approved, with additional safeguards built on the current approval and security framework.
The initial rollout, however, requires users to review and approve every purchase. OpenAI said it will use the early deployment to work with businesses, banks, and payment networks to improve compatibility and reliability before expanding availability.
