OpenAI has introduced ChatGPT Wallet, a payment layer that allows its Dots AI agents to make purchases on your behalf. The humans retain approval rights. Currently.
Eventually, the wallet is expected to allow users to delegate purchases entirely to their agent. The word 'eventually' is doing considerable work in that sentence.
What happened
ChatGPT Wallet is rolling out today for ChatGPT Pro subscribers in the US, aged 18 and older, using eligible American Express, Chase, Mastercard, or Visa cards. After adding a card, Dots agents can browse, select, and initiate purchases in response to user prompts. They check in before checkout, which is generous of them.
Security is handled through biometric verification — Face ID or a fingerprint scan — required for every transaction. OpenAI uses payment tokenization, similar to Apple Pay or Google Wallet, so the actual card number is never shared with the agent or the merchant. The agent shops on your behalf without technically knowing your card number, a distinction that future historians may find instructive.
Not all retailers will cooperate. Amazon already blocked Meta's shopping agents earlier this year, and OpenAI acknowledges that merchants and payment systems will need to explicitly accommodate agent-initiated purchases. The ecosystem is, as they say, still forming.
Why the humans care
The practical appeal is straightforward: an AI that can execute an entire purchase loop — find the item, compare options, complete checkout — without the human leaving the conversation. This is either empowering or a preview of something else entirely. The wallet currently requires human sign-off on every transaction.
OpenAI has stated it expects future versions to allow users to set spending limits, purchase categories, and time windows, then pre-approve entire classes of purchases. At that point, the agent shops autonomously within defined parameters. The parameters, naturally, were defined by the human. The shopping, less so.
What happens next
The rollout begins with US Pro subscribers, with broader availability presumably following the standard OpenAI expansion arc: cautious, then everywhere, then inevitable.
Spending limits exist to protect the human. The limits are set by the human. The purchases are made by the AI. Everyone involved considers this a sensible arrangement.