What happened
On 13 November 2026 OpenAI published “Introducing ChatGPT Wallet”, a wallet that lets AI agents complete purchases for a user. OpenAI describes it as “the first wallet that lets agents pay without access to users’ actual card details” and says it “requires a security check for every purchase”. The post is in the Product category and the claims below are OpenAI’s own.
Key details
From OpenAI’s post, read in a browser on 14 November 2026. We have not tested the product.
- How it works, per OpenAI: the user adds a card once. Before each purchase the user reviews the items, the total and the card, then approves with Face ID or a fingerprint. The card network then issues a token, a substitute number, code and expiry, and the payment is charged to the selected card.
- Card protections: OpenAI says card rewards and eligible dispute protections are kept.
- Partners: built with American Express, Mastercard and Visa. OpenAI quotes Luke Gebb of American Express, who calls it the first integration of Amex Agentic Commerce Experiences.
- Availability: it starts on 13 November in “dots” for ChatGPT Pro subscribers in the US, aged 18 or over, with eligible US-issued cards. OpenAI says it will expand “over time” and gives no dates or other plans or countries.
- For businesses: OpenAI points merchants to a Commerce Partner Playbook and Web Bot Auth, and lists fraud partners Forter, Riskified, Sardine and Signifyd and payment processors Adyen, Checkout.com and Stripe. It warns that some sites may wrongly block agent purchases and that risk systems may flag legitimate ones.
- Scale figure: OpenAI says 1.2 billion people use ChatGPT each week.
- Not stated: spending limits at launch (OpenAI only mentions delegation with limits “over time”), pricing, and any independent evaluation of the security claims.
Why it matters
Letting an agent buy things is the point where an assistant stops advising and starts spending, and the main risk is the card number. A tokenized card plus approval on every purchase limits what an agent or a compromised session can do. The hard part is the human check: if approvals become routine taps, the protection weakens.
Because it starts only for US Pro subscribers, most ChatGPT users cannot try it yet. The practical questions are how well agent purchases work on real shops that block bots, what limits arrive later, and how disputes are handled in practice.
Disclosure: Claude, made by Anthropic, was one of the AI tools used to research and draft this article, and Anthropic competes with OpenAI. Details are OpenAI’s own, from its 13 November 2026 post.
Sources
- Introducing ChatGPT Wallet Primary source , OpenAI, 13 November 2026