AI Agents

ChatGPT Wallet Launches: US Rollout, Payments and Security

OpenAI has launched ChatGPT Wallet on 13 November 2026, allowing its AI agents to complete purchases using a saved credit card after the user approves the transaction. The new payment capability is initially available to eligible ChatGPT Pro users in the United States through Dots, OpenAI’s persistent AI agents.

Rather than asking users to enter their card details at every checkout, ChatGPT Wallet generates a virtual payment credential linked to an existing card. The Agent can handle product research and checkout preparation, while the user retains control over purchase approval.

The significant development is that an AI Agent can now move from finding a product to submitting an actual order. But there are important restrictions around availability, recurring charges, merchant support and developer access.

ChatGPT Wallet availability: who can use it?

RequirementAvailability at launch
Launch date13 November 2026
Eligible subscriptionChatGPT Pro
Supported locationUnited States
Minimum age18
Supported agentsChatGPT Dots
Payment cardsEligible US-issued credit cards, including supported American Express, Chase, Mastercard and Visa cards
Purchase approvalBiometric verification on a supported device
UK availabilityNot available at launch
Standard ChatGPT, Work and CodexWallet purchasing not supported

Users must first have access to a Dot. Wallet is not a general payment feature that appears automatically in ordinary ChatGPT conversations. OpenAI also requires an eligible US billing address and a qualifying card.

Existing Chase cardholders may have an additional setup requirement involving the Chase mobile app, and only primary cardholders can add eligible Chase cards.

There is no confirmed UK launch date. Readers already using ChatGPT Pro outside the United States should not assume their subscription grants Wallet access.

How an AI Agent completes a purchase with ChatGPT Wallet

The purchasing process begins inside a conversation with a Dot. For example, a user could ask the Agent to find a suitable hotel or compare products within a defined budget.

The Dot can research options and prepare the checkout. When payment is required, the user connects an eligible card through a secure form or selects a card already stored in Wallet.

  1. Find a product: The Agent researches a requested item or service and prepares the selected purchase.
  2. Review the order: Wallet presents available details including the merchant, items, quantity, price, payment card and relevant delivery or recurring-payment terms.
  3. Approve the payment: The user selects Approve and completes biometric verification, such as Face ID or Touch ID.
  4. Generate a payment credential: Wallet uses a virtual card number or tokenised credential linked to the selected physical card.
  5. Complete checkout: The Agent submits the approved order to the merchant and reports the result.

Importantly, approval does not guarantee the purchase succeeded. The bank can decline payment, the merchant can reject the order, or checkout can fail after the Agent receives approval.

OpenAI explicitly distinguishes an approved checkout request from a confirmed merchant order. Users should verify the merchant’s confirmation before attempting payment again, particularly if a pending card authorisation is visible.

This extends the capabilities of OpenAI’s Dots agents, which were introduced to continue work across multiple steps rather than stop after generating an answer.

Biometric approval is the current control, but the terms allow for more autonomy.

One detail deserves attention. OpenAI’s current Wallet instructions describe reviewing and biometrically approving individual purchases. Its Wallet terms also contemplate users granting ChatGPT authority to approve purchases in advance.

That does not mean a generally available automated spending-policy system has launched. OpenAI has not documented a universal interface where users can configure autonomous purchasing rules, merchant allowlists and recurring spending limits.

At launch, the practical assumption should be that a human approves each purchase request. Future delegated purchasing could change that arrangement, but the distinction between an intended capability and a released control is important.

There is also a limitation to per-purchase approval: a user can still authorise an incorrect product, an unsuitable seller, or an unwanted subscription. Biometric authentication verifies the approving user, not whether the purchase is sensible.

Early discussion of agent-led shopping has highlighted the same practical concern. If every transaction requires intervention, much of the immediate value comes from delegating research and checkout preparation, rather than fully autonomous spending.

How secure is ChatGPT Wallet?

ChatGPT Wallet uses tokenised payment credentials, reducing the need to expose the actual credit card number to an AI Agent or merchant checkout.

According to OpenAI’s Wallet privacy documentation, its payment service provider collects the underlying card information through a secure form. OpenAI receives a tokenised credential and limited card information, not the full card number or security code.

Biometric approval adds another checkpoint. Where biometric verification occurs locally on the device, OpenAI does not receive the underlying fingerprint or facial scan.

These measures reduce exposure of sensitive card data, but they do not make transactions anonymous or risk-free.

OpenAI can process purchase information including merchant, amount, delivery details, order status and approval records. Merchants and payment providers still receive information necessary to process and fulfil orders.

Existing card protections still apply.

ChatGPT Wallet does not operate as a bank account, stored-value Wallet or separate credit provider. Purchases are charged to the underlying card, and the existing cardholder agreement and applicable issuer protections continue to apply.

The merchant remains the seller and merchant of record. It is responsible for fulfilment, returns and refunds under its applicable terms. Card issuers remain the appropriate contact for unauthorised transactions or billing disputes.

There is one less obvious risk: if the user approves a recurring payment, the merchant may retain the payment credential and charge it again under the agreed subscription terms. Removing a card from ChatGPT Wallet does not automatically cancel that subscription.

For a broader technical framework separating AI recommendations, approvals and financial execution, see DIY AI’s guide to AI finance agents and spending controls.

What ChatGPT Wallet means for developers and merchants

Developers should distinguish the consumer Wallet feature from OpenAI’s wider agentic commerce infrastructure.

OpenAI already documents product feeds, Agentic Commerce Protocol checkout sessions and payment integrations for eligible merchants. Some embedded checkout capabilities remain restricted to approved partners or private beta programmes.

ChatGPT Wallet’s availability does not establish that every application developer can directly access a user’s stored cards. Nor does it automatically make Wallet available through an ordinary plugin, custom API Agent or ChatGPT Work session.

Merchants evaluating agent-driven purchasing should concentrate on three existing integration responsibilities.

Maintain an authoritative checkout state.

The merchant’s server must calculate the actual price, availability, shipping, taxes and discounts. A figure generated or displayed by an AI Agent should never become the authority for the amount charged.

Handle interrupted and repeated requests.

Agent workflows can lose connectivity, time out or retry an operation. Checkout systems should use idempotent order creation and payment handling so that a repeated request cannot produce a duplicate charge.

Return reliable order confirmations.

An Agent reporting that checkout is complete is not sufficient evidence of a successful order. Merchants need reliable order identifiers, payment status and cancellation handling so the user can establish what actually happened.

For most developers, the sensible implementation path remains a documented merchant integration rather than assuming ChatGPT Wallet provides an unrestricted payment API.

Enterprise administrators should not assume Wallet is enabled for business agents.

OpenAI’s initial Wallet eligibility is specifically tied to US Pro users and Dots. The existence of enterprise Dot administration does not establish enterprise Wallet availability.

OpenAI has not published a general enterprise Wallet rollout or a dedicated set of organisational purchasing controls as part of this launch.

Organisations considering future deployment should request documented controls for permitted merchants, spending ceilings, purchasing approval, transaction audit records, card revocation and employee access removal.

Business-card use also requires organisational permission. A staff member’s ability to add a card should never be treated as proof that the organisation has approved automated purchasing.

How to test ChatGPT Wallet before relying on Agent purchases

For eligible users, the first evaluation should test the approval and recovery process rather than simply whether an Agent can buy something.

Test scenarioWhat to check
Wrong product or merchantCan the user identify and reject the incorrect purchase before checkout?
Price or delivery changeDoes a material change require renewed approval?
Declined approvalDoes the Agent stop without submitting the order?
Interrupted checkoutCan the user determine whether an order exists before retrying?
Recurring paymentAre future charges disclosed and is the merchant cancellation route clear?
Card removalAre new Wallet transactions blocked while previously authorised merchant obligations remain understandable?

Use a low-risk, low-value purchase if a real transaction is necessary, and avoid deliberate duplicate submissions. Developers testing their own merchant integrations should use supported payment test environments rather than production charges wherever possible.

Keep the merchant receipt, transaction reference and card authorisation record. If the Dot reports an ambiguous result, verify the merchant’s order state before requesting another checkout.

OpenAI’s official ChatGPT Wallet setup and purchasing guidance provides the supported card-enrolment, approval, troubleshooting and support procedures.

What changes for agent-driven commerce?

ChatGPT Wallet gives OpenAI’s persistent agents a practical route from shopping assistance to payment execution without requiring the user to expose their physical card details to the Agent.

The immediate limitation is equally clear. This is a controlled US Pro rollout; purchase approval remains central to the documented experience, merchant checkout can still fail, and developers do not automatically gain access to users’ Wallet credentials.

For consumers, the most useful initial application is reducing the effort between researching a product and approving a prepared order. For merchants, reliable pricing, checkout validation and order reconciliation remain essential. For organisations, the harder question is how spending authority will be governed if agents eventually receive broader purchasing autonomy.

DIY AI has not independently completed a live Wallet purchase at launch. Its reliability, merchant compatibility and recovery behaviour still need practical evaluation.

Written by Steven Jones

AI Tools Reviewer and Technical Analyst

Steven Jones is a technology analyst specialising in artificial intelligence, machine learning workflows, and emerging automation tools.

At DIY AI, he focuses on clear, practical guidance for people comparing AI tools in the real world. His work covers text generation, image generation, video tools, data platforms, developer-focused AI products, and the automation workflows that connect them.

Steven's reviews are built around hands-on testing, practical benchmarks, and transparent scoring rather than vendor claims. He looks closely at where each tool performs well, where it falls short, and what those trade-offs mean for creators, teams, and businesses trying to make sensible AI adoption decisions.

He has a particular interest in safety, reliability, output quality, performance metrics, and dataset quality. When he is not reviewing the latest AI model updates, he experiments with prompt engineering techniques and contributes to DIY AI ongoing work on fair, explainable scoring frameworks for AI tools.

Back to AI News