OpenAI Launches ChatGPT Wallet for Agent-Driven Checkout

By Amara Osei

On Friday, OpenAI launched a wallet feature that lets its AI agents — called dots — complete purchases on a user's behalf, opening the first controlled test of agentic checkout inside ChatGPT. The rollout starts small: a limited number of U.S. ChatGPT Pro subscribers, aged 18 and up, holding a U.S.-issued Visa, Mastercard, or American Express card.

The mechanism matters more than the announcement. Understanding how money moves here explains both the appeal and the risk.

How does the payment flow work?

Subscribers add an eligible card to the ChatGPT Wallet once — no re-entry of payment details on later purchases. They then ask their dot to help with a purchase. At checkout, the wallet charges the selected card using a payment token, a substitute for the actual card number. That token means the real card details are never shared with either the dot or the merchant.

The merchant automatically receives what it needs to fulfill the order from the dot — shipping address and contact information, according to OpenAI.

What control do users keep?

OpenAI frames the user as the final gate in the transaction. Before checkout, users review the items, the total, and the payment card. Each purchase then requires approval through a biometric security check such as Face ID or a fingerprint, which also creates a transaction record.

Card benefits carry over. OpenAI says shoppers using dots still earn credit card rewards and keep standard card protections, including the ability to dispute transactions.

The one-line takeaway: tokenization plus biometric approval is the safety architecture OpenAI is betting on.

Why is OpenAI moving into agentic shopping now?

The market signals are mixed but growing. DoorDash and Uber have reported low order volumes from AI agents so far. Meta's Muse, with more than 10 million users, has seen AI-assisted shopping become one of its top activities. Other personal agents, such as Instinct, already support purchases on a user's behalf.

That divergence — weak ordering on delivery platforms, strong engagement on general assistants — is the gap OpenAI intends to occupy.

What comes next?

The limited launch lets OpenAI test and refine the feature before a broader rollout. Planned additions include:

• Spending limits set by the user
• Pre-approval parameters by purchase category or time period
• More autonomous purchase handling by the agent once pre-approvals are configured

OpenAI also wants the feature available at a wider set of retailers. That is not guaranteed: some retailers actively block AI agents from their sites. The largest is Amazon, which bans agents in favor of steering users to its own assistant, Alexa.

To ease retailer adoption, OpenAI published a Commerce Partner Playbook with tools and guidance for sites that want to support agent-driven customers.

What does this mean for enterprises?

For CIOs and commerce leaders, the launch sets a concrete reference architecture: tokenized payment credentials, human biometric approval, and a merchant integration playbook. Retailers that block agents outright — as Amazon does — versus those that follow OpenAI's playbook will define the next split in agentic commerce.

The question for merchants is no longer whether agents will arrive at checkout, but on whose terms.