deniz.in

Markets

Weather

Loading weather

· via dev.to (home feed)

OpenAI launches ChatGPT Wallet so AI agents can pay, with tokenized cards and biometric approval

OpenAI's ChatGPT Wallet lets its always-on 'dot' agents complete purchases using tokenized card credentials, with a biometric check required for every transaction. Access starts with US ChatGPT Pro subscribers.

OpenAI launches ChatGPT Wallet so AI agents can pay, with tokenized cards and biometric approval

What OpenAI launched

On Friday, 13 November 2026, OpenAI announced ChatGPT Wallet, a payment feature that lets its AI agents complete purchases on a user's behalf. According to OpenAI, the wallet allows an agent to pay while never gaining access to the user's real card details, and it requires a security check on every transaction. The feature was built in partnership with American Express, Mastercard and Visa, and is initially available to ChatGPT Pro subscribers in the United States through "dots", which Reuters describes as OpenAI's always-on AI agents. A post on dev.to, working through the announcement alongside OpenAI's help center documentation, lays out the mechanics and the fine print.

How a purchase works

The user adds a card once. Before each purchase, they review the items, the total and the card being charged, then approve using a security check such as Face ID or a fingerprint scan. Once approved, the payment network issues a payment token — a stand-in credential with its own card number, security code and expiration date — and that token is what the merchant receives. Neither the merchant nor the agent sees the genuine card number, and OpenAI notes that the number appearing at checkout may not match the physical card as a result. Each approval also creates a record of the purchase the user authorized the agent to make, which functions as a built-in audit trail.

Eligibility and availability

The launch is deliberately narrow. Users must be 18 or older, live in the United States, and hold a US billing address. Only US-issued credit cards from American Express, Chase, Mastercard or Visa are supported; debit cards are excluded, and virtual cards are not supported at launch. The wallet works only in conversations with a dot, not in regular ChatGPT sessions, Work accounts or standalone Codex threads. There is one unusual restriction around Chase: only the primary cardholder can add a Chase card, not an authorized user. Because the rollout is gradual, even users who meet every requirement may not have access yet.

What happens when something goes wrong

OpenAI's help center splits responsibility three ways. Merchants handle order status, cancellations, returns and refunds. Card issuers handle declined, pending, disputed or unrecognized charges, plus bank verification. OpenAI Support handles wallet errors, suspected unauthorized access to a ChatGPT account, and purchases that differ from what the user actually approved.

Two caveats stand out. A "ready to check out" status means the user's approval is complete, but it does not confirm the merchant has placed the order — OpenAI advises checking for the merchant's confirmation before retrying. And if the user agrees to a recurring payment plan, the merchant may store the payment credential and charge it again under the agreed terms, with cancellation handled between the user and the merchant. OpenAI also warns users never to share full card numbers, security codes, passwords or bank verification codes inside the chat or in support messages; card details belong only in the secure form.

The merchant side of the equation

For this to work at scale, the rest of the web has to cooperate. OpenAI says websites need to avoid accidentally blocking agents, and that payment and risk systems need to recognize legitimate purchases made by agents. Alongside the wallet, the company published a Commerce Partner Playbook covering product information and feeds, signed requests via Web Bot Auth so CDNs and firewalls do not block ChatGPT agents, and coordination with payment and risk providers to reduce unnecessary declines. OpenAI lists fraud-prevention partners including Forter, Riskified, Sardine and Signifyd, and payment providers including Adyen, Checkout.com and Stripe. Even so, the help center acknowledges the wallet may not complete checkout at every store. OpenAI puts its own scale at 1.2 billion weekly ChatGPT users, many of whom already use the product to research and compare purchases.

Where OpenAI says this is heading

Today's design is conservative: nothing ships without a human approving it. But OpenAI says it expects users to eventually delegate purchases entirely to their agents by setting parameters such as spending limits, purchase categories or time periods, and pre-approving that activity with safeguards that build on the current protections, including biometric checks. That second stage — autonomous agent spending within pre-set bounds — is where agent-driven commerce would shift from convenience to infrastructure.

Why it matters

This is one of the first large-scale attempts to give AI agents a payments rail backed by the major card networks. The design addresses the two obvious risks — card detail exposure and unauthorized spending — through tokenization and per-transaction biometric approval, and the authorization records add an audit trail that has been missing from most agent demos. The stated roadmap toward fully delegated spending raises harder questions about disputes, liability and how limits are enforced when no human reviews a transaction. It also inverts a long-standing assumption on the web: for two decades, infrastructure treated bot-like traffic as a threat to block, and now merchants and fraud systems must distinguish an approved agent from an abusive one — those that cannot will decline legitimate sales. The rollout is small, but the direction is clear: agents are becoming parties at checkout, and the norms around verifying who is actually paying will have to change with them.

  • #ai-agents
  • #payments
  • #openai
  • #chatgpt
  • #fintech

Related posts