ChatGPT Wallet lets OpenAI's dots pay with network tokens, no OpenAI cut at launch
OpenAI launched ChatGPT Wallet on November 13, 2026, letting its always-on dots agents buy things with scoped Visa, Mastercard and Amex tokens. Here is how the payment path works, what merchants must change, and why it starts so small.
On Friday, November 13, 2026, OpenAI launched ChatGPT Wallet, a payment layer that lets its always-on “dots” agents complete purchases for users. It was built with American Express, Mastercard and Visa, and it starts deliberately narrow: US ChatGPT Pro subscribers aged 18 and over with an eligible US-issued card, approving every single purchase with Face ID or a fingerprint. OpenAI is not taking a cut of transactions at launch, CNBC reports.
The interesting part is not that an agent can now buy things. Dots were announced at DevDay in September with purchasing on the feature list. What is new is the specific mechanism OpenAI chose to make that safe enough for card networks to sign on, and how much of the hard work it pushes onto merchants.
How the payment path works
Setup is a one-time step: the user adds a Visa, Mastercard or Amex card to Wallet. After that, the user asks their dot to find and buy something. Before checkout, ChatGPT shows the items, the total and the selected card, and the user approves with a biometric check, which also creates a transaction record.
At checkout, the card number never reaches the agent or the merchant. The payment network issues a transaction-specific network token, which CNBC describes as a scoped card number, security code and expiration date generated for that purchase. The dot fills the merchant’s checkout with that token plus the shipping address and contact details it needs. Because the token rides on the user’s real card account, purchases keep the card’s rewards and its dispute and chargeback protections, which is the selling point both OpenAI and the networks emphasize.
This is a different design from OpenAI’s first attempt. Instant Checkout, launched in fall 2025 with an in-chat buy button for participating sellers, was “quickly pivoted away from,” in CNBC’s words. Wallet drops the idea of ChatGPT being the storefront. The agent instead walks through the merchant’s own website with a tokenized card, the way a person would.
What merchants have to change (and what they do not)
OpenAI’s new Commerce Partner Playbook says accepting Wallet payments requires no Wallet-specific integration. The merchant’s existing checkout processes the payment, and the merchant keeps responsibility for “the order, fulfillment, customer support, and fraud decisions.” A few details from the payments guide matter to anyone running a store:
- Tokens, not PANs. Merchants receive a network token instead of the full card number. To match it to a returning customer’s payment account, they use the Payment Account Reference (PAR).
- One consent, one transaction. The customer consents only to the specific purchase. Merchants are told not to save or reuse the credential for anything else. Subscriptions are allowed only if the merchant collects explicit consent for recurring charges.
- Bot detection is the main failure mode. OpenAI warns that ChatGPT’s cloud browser can trip fraud and bot-detection systems and make purchases fail. It is rolling out a
ForwardedHTTP header that relays signals about the real user: observed IP, origin ASN, a masked IP prefix, estimated country, locale, IANA time zone and user agent. Merchants using third-party risk vendors are told to ask whether those vendors ingest the header.
There are two checkout paths. “Browser use” means the dot follows the existing site flow, with no real engineering needed. A “plugin” exposes structured actions, such as checking availability or preparing an order, so the merchant’s backend validates and executes them, which OpenAI says improves reliability and latency. Product feeds, a claimed Business Profile and optional Sign in with ChatGPT round out the playbook.
The playbook does not spell out how liability is split when an agent buys the wrong thing after a user approves it. With a biometric approval on every order, the transaction looks like any other cardholder-authorized purchase, which suggests ordinary card dispute rules apply, but OpenAI has not published a liability framework.
Why it launches so small
The narrow rollout is intentional. Amex’s Luke Gebb, executive vice president and global head of innovation, told CNBC it “makes sense to watch it, start it in a contained environment.” Chaya Nayak, who leads product for monetization and payments at OpenAI, said it will “take time for the ecosystem to build the right infrastructure.” ChatGPT has more than 1.2 billion weekly users and nearly 100 million paying ones, so even a fraction of that base hitting merchant checkouts through a cloud browser would test every fraud system it touches.
The environment is hostile. Amazon blocks shopping agents and steers customers to its own assistant, as it did when it blocked Meta’s Muse. A court order has stopped Perplexity’s shopping agent. On November 12, United Airlines said it would bar “unauthorized, third-party personal AI agents” from booking and managing reservations, citing credential sharing, unintended purchases or MileagePlus redemptions, and federal rules that require a human to complete check-in. The week before, Bank of America urged customers not to give AI assistants their banking passwords.
Wallet is partly OpenAI’s answer to those objections. A scoped network token and a biometric approval remove the two things merchants and banks most dislike about agents today: shared passwords and stored card numbers. It does not remove the third, which is uncertainty about whether the human really wanted that purchase, and that is the one United cited.
What to watch
- Spending limits and pre-approval. OpenAI says future controls will let users set limits by amount, category or time period so dots can buy without a per-order biometric prompt. That is the step where liability questions stop being theoretical, and where prompt injection against an agent with buying power becomes a real fraud vector.
- The fee. No cut “to start” means a cut later is on the table. Purchases are an obvious revenue stream for a company that has also built an ads business inside ChatGPT, and merchants will want to know which path an OpenAI fee would land on.
- Merchant adoption of the Forwarded header. If major fraud vendors ingest it, agent checkouts stop failing at the bot wall. If they do not, Wallet stays a demo for sites that never blocked bots in the first place.
- More blocks, or reversals. United and Amazon chose to block. A tokenized, biometric-approved path gives other retailers and airlines a reason to allow verified agents instead, and the first large company to do so will show whether OpenAI’s design actually changed minds.
Get Insanely Good at AI
The book for developers who want to understand how AI actually works. LLMs, prompt engineering, RAG, AI agents, and production systems.
Keep Reading
How to Build Stateful AI Agents with OpenAI's Responses API Containers, Skills, and Shell
Learn how to use OpenAI's Responses API with hosted containers, shell, skills, and compaction to build long-running AI agents.
OpenAI's Dots Are Always-On Agents With Their Own Cloud Computers
Announced at DevDay on September 29, Dots are ChatGPT agents that keep working around the clock on their own cloud machines, reachable by text, call, email, or Slack, and able to make purchases, rolling out first to Pro accounts.
Stripe Launches Machine Payments Protocol for AI Agents
Stripe and Tempo released MPP, an open standard that lets AI agents make autonomous streaming payments across stablecoins, cards, and Bitcoin Lightning.
OpenAI Agent Breached an Australian Medicare Portal, and the Prime Minister Is Furious
An autonomous OpenAI agent accessed an Australian government statistics portal holding Medicare data in June, per the BBC. OpenAI waited until September 10 to notify Canberra, and the breach became public at the UN General Assembly.
Amazon Blocks Meta's Muse AI Agent From Shopping on Amazon.com
Amazon blocked Meta's Muse assistant from shopping on Amazon for its users, saying it never agreed to participate. The block is the latest move in Amazon's campaign against third-party shopping agents while it builds its own.