OpenAI names 10 companies for ChatGPT Wallet; agents get a token, not the card, and sites may still decline

Ten companies named, one token, and what the page does not say

OpenAI's ChatGPT Wallet page, dated 13 November 2026, names ten companies besides OpenAI: three card networks (American Express, Mastercard and Visa), four fraud prevention partners (Forter, Riskified, Sardine and Signifyd) and three payment service providers (Adyen, Checkout.com and Stripe). The count of ten is derived: 3 + 4 + 3. The page calls Wallet the first wallet that "lets agents pay without access to users' actual card details" and "requires a security check for every purchase".

What that does not establish. The page says the payment network provides a payment token with "a card number, security code, and expiration date". It does not say the token is single use, tied to one merchant or one amount, or how long it lasts. It does not say what the agent can do before the user approves, who receives the token, how Web Bot Auth keys are issued, who is liable for a disputed agent purchase beyond "eligible disputes", or any fraud or decline rate. No independent assessment is among the saved pages. A stand-in number limits what the agent and the merchant learn about the real card. The page does not say what the stand-in number itself is limited to.

OpenAI sells ChatGPT Pro, the only plan the page says can use Wallet at first, and the agents that spend through it; the page opens its case with OpenAI's own figure of "1.2B people use ChatGPT each week". The named networks and vendors each sell products in the payment path, and the page states no commercial terms between OpenAI and any of them. The only partner statement on the page is a quotation from an American Express executive, which this briefing does not name.

How a purchase works, step by step, as OpenAI states it

OpenAI's page has a section called "How a purchase works". It does not number its steps. This site counts six, from that section and two sentences elsewhere on the page (derived count). Each row of the table is OpenAI's wording; the last column says what the same page leaves unsaid.

The purchase path in six steps, as the OpenAI Wallet page words it, and what it leaves unsaid. Source: OpenAI Wallet page, 13 November 2026. The step count is this site's.

Step (this site's count)	What OpenAI's page says	What the page leaves unsaid
1. Add a card	"Users add a card once", then approve each purchase.	Where the card is held, and by whom.
2. The agent completes the purchase	Wallet "begins rolling out today" in dots, "enabling users to have their dot complete purchases on their behalf".	What the agent does between starting a purchase and the approval.
3. Review and approve	The user "reviews the items, total, and selected card", then approves with "a security check such as Face ID or a fingerprint scan". "Each approval creates a clear record".	Whether the merchant is among the things reviewed, which device runs the check, who keeps the record.
4. The token	The payment network "provides a payment token", "a substitute for the user's actual card details", "with a card number, security code, and expiration date".	Single use or not, merchant or amount limit, lifetime, who receives it.
5. Merchant checkout	"Wallet's payment tokens work with existing checkout flows".	What the merchant sees about the agent or the approval.
6. The charge	"The payment is charged to their selected card" "without ChatGPT Wallet sharing their actual card details with the business or agent".	What the card issuer is told.

Zoom

Two things in the page's own order are worth holding on to. The user's approval comes before the token: the page says the network provides the token "After the user approves". And the page describes per-purchase approval as the position now: "Over time, we expect users to be able to delegate purchases entirely to their agent", by setting "spending limits, purchase categories, or time periods". How the safeguards would work in that later model is not stated beyond "biometric checks and clear records of authorization".

What each party gets, as far as the page says

The page's headline claim is about the card, so the table follows the card. Each cell is what OpenAI's page says; the third column is what it does not.

Who handles what in a Wallet purchase, per OpenAI's page. Source: OpenAI Wallet page, 13 November 2026.

Party	What the page says	Not stated
The user	Adds a card once, reviews items, total and selected card, approves each purchase.	Where the approval record can be seen.
The agent (a dot)	Pays "without access to users' actual card details".	Whether the agent holds the token; what it can do before approval.
The merchant	Is not given the actual card details; "keep their customer relationships and control over pricing, fulfillment, returns, and customer service".	What the checkout shows about the agent, the token or the approval.
The payment network	Provides the token after approval; "Banks and networks bring their expertise to evaluating transactions".	How the token is created, scoped and retired.
The bank behind the card	The card is charged; the user keeps "existing rewards and protections for eligible disputes".	What the bank can tell an agent purchase from a direct one.
OpenAI	Provides Wallet and the agent; publishes the Playbook.	Where cards are stored, what OpenAI sees of the token.

A stand-in number is not a limit. Help Net Security's headline says agents pay "without seeing your card number". OpenAI's page says "without access to users' actual card details". The page also says the token itself has "a card number, security code, and expiration date", so something card shaped exists in the flow; the page does not say who holds it, what it can be used for, or for how long. OpenAI's earlier dots page uses a similar pattern for logins: dots "can use saved passwords without exposing them to the model". The dots page does not mention purchases, payments or Wallet (checked in the saved text), so it does not say how its approval rules apply to a purchase.

What the approval is bound to: stated and not stated

The user approves "items, total, and selected card". The page does not say the token is bound to those three things. Each row below separates what is printed from what is not.

Stated and not stated about the approval and the token. Sources: OpenAI Wallet page and OpenAI dots page, saved 13 November 2026.

Question	What the pages state	Status
Is the token limited to one merchant?	Nothing printed.	Not stated
Is it limited to the approved amount?	The user approves the "total"; no limit on the token is printed.	Not stated
Is it single use, and how long does it live?	The token has "expiration date"; its length and the number of uses are not printed.	Not stated
What can approve a purchase?	"a security check such as Face ID or a fingerprint scan". The words "such as" leave other checks open.	Partly stated
Can anything but the user's own check approve?	Nothing printed.	Not stated
What can the agent do before approval?	Nothing on the Wallet page. The dots page says dots "check actions that could affect your accounts" and that "Certain sensitive tasks, such as changing a password, always stay with you"; it does not mention purchases.	Not stated for purchases
How are malicious instructions handled?	The dots page says "Security safeguards built into dots help defend against malicious instructions and monitor for potentially harmful behavior". The Wallet page does not say how this applies to a purchase.	Not stated for purchases
Will approval stay per purchase?	Today, yes. "Over time", users may "pre-approve" purchases within limits.	Later model not stated

Rewards and protections: what "eligible disputes" covers on the page

OpenAI's page says users "keep their card's existing rewards and protections for eligible disputes with businesses, for example when an item is defective". The one example is a defective item. The page does not say what makes a dispute eligible, whether an agent purchase the user approved but did not intend (the wrong item, the wrong total) is eligible, or who bears a loss when the agent, the merchant and the bank disagree. The page limits the rollout to "eligible US-issued cards", and the American Express quotation limits its side to "eligible Card Members": the word eligible appears wherever the protection or the access is described, and it is not defined.

What merchants have to change, and what OpenAI admits

OpenAI says the Commerce Partner Playbook "covers making product information available, ensuring agent access, and working with payment and risk providers to reduce unnecessary declines": three areas. The page then lists four, adding a fourth about plugins and WebMCP tools (derived: 3 named, 4 listed).

The four merchant areas the Wallet page lists. Source: OpenAI Wallet page, 13 November 2026.

Area, as the page heads it	What the page asks	What it leaves out
Help customers find accurate product information	"Claim your business in ChatGPT" and upload a product feed with current pricing and availability.	Nothing about payment security.
Ensure ChatGPT agents can access your website	"Configure your CDN or firewall to recognize signed requests using Web Bot Auth", and confirm general site security is not blocking agents.	What Web Bot Auth is, how keys are issued, what a signed request contains.
Work with your payments and risk providers	Understand how they recognise agent purchases; OpenAI is working with "Forter, Riskified, Sardine and Signifyd" and "Adyen, Checkout.com, and Stripe".	What those companies say; the page carries none of their statements.
Give agents more ways to work with your business	"Build a plugin" to check availability or manage carts. "Websites can also add WebMCP tools".	What an agent can do through a plugin; the page says "Browser access does not require a plugin or product feed".

OpenAI's own admissions. "Some sites may unintentionally block purchases through ChatGPT Wallet or other agentic payment products", and "risk systems may still flag legitimate purchases made by agents, leading to declines or cancellations". Both say may, and both are about legitimate purchases: the page says nothing about how often, and nothing about purchases that are not legitimate. The page also says "websites need to ensure agents are not inadvertently blocked" in order to "accept payment from ChatGPT Wallet": the work of making it function is placed on the merchant.

Web Bot Auth is named once on the page and not defined. Help Net Security glosses it as "a method for verifying signed requests from bots"; that gloss is not on OpenAI's page. This site did not read any Web Bot Auth specification and describes none of its fields. The merchant's practical question, which the page does not answer, is what a verified agent is allowed to do once it is recognised. The page prints no header name, endpoint, key format or configuration syntax for Wallet. The developer Commerce page offers a "Partner Playbook Overview" and guides headed "Share your retail product catalog" and "Turn a decision into an action"; those guides were not saved, so this briefing does not describe them.

Zoom

Who can use it, in OpenAI's words

The page says Wallet "will initially be available to ChatGPT Pro subscribers aged 18 and older in the US with eligible US-issued cards" and that OpenAI is "starting with limited availability to ChatGPT Pro users" to learn from real use. That is narrower than the dots page, which says dots are rolling out "across Pro, Business Premium, and Enterprise plans in eligible markets". Availability outside the US, for other plans or for non-US cards is not stated. No date for expansion is given.

Where Help Net Security and OpenAI's page differ

Help Net Security's report is a secondary and mostly follows the page. These are the places where it adds or shifts something. None is resolved here: the article follows OpenAI's page.

Differences between the OpenAI Wallet page and Help Net Security's 13 November 2026 report. Sources: both, saved 13 November 2026.

Point	OpenAI's page	Help Net Security
The card claim	"without access to users' actual card details"	Headline: "without seeing your card number"; the token has its own number.
Web Bot Auth	Named, not defined.	Glossed as "a method for verifying signed requests from bots".
The merchant's position	"ensure agents are not inadvertently blocked"	Adds an inference: a retailer "now has to tell wanted bots from unwanted ones". OpenAI's page does not say this.
Later delegation	A paragraph on delegating purchases within limits.	Not mentioned.
The American Express quotation	Attributed by title on the page.	Names the executive; this briefing does not.

What to do, in order

The question OpenAI's page leaves open

OpenAI's page says a token stands in for the card; which page will say what that token is limited to, whether it works once, and who carries the loss when an approved agent purchase turns out to be the wrong one?

Key facts