American Express publishes an intent framework to bind AI agent purchases to what the customer approved

American Express published its Intent Intelligence framework on 13 October 2026, setting out how it thinks a customer's purchase intent should be captured, authenticated and kept attached to a transaction when an AI agent shops on that customer's behalf. Amex said it is making the design public in the hope of a shared industry approach, and presented it the same day at TechCrunch Disrupt. The model has four steps. A Create Intent API records the shopper's instructions and limits, such as the merchant, a maximum budget and product preferences. The customer then verifies their identity and approves that intent. Amex issues an Intent ID and a Proof of Intent Token describing the constraints, and the agent pays with a scoped digital credential derived from that token rather than the real card number. The Intent ID and token can then be checked later in the flow to confirm a purchase matches what was authorised, and to inform fraud screening, authorisation and dispute handling. Amex says the main difference from other approaches is that the issuer sees the approved intent before the agent completes the purchase. In its own example, an agent told to buy one brand of trainers in a given size could be blocked if it tried to buy another brand. The framework is a core service in the ACE developer kit Amex launched in April, and the company says it is working with EMVCo and the x402 Foundation. It named no merchant partners, gave no rollout date and said it has not yet seen a rise in agent-related disputes. For online merchants, the framework signals how at least one network expects agent orders to be judged: against an explicit, issuer-held record of what the cardholder asked for. Orders that drift from that record, on merchant, amount or item, are the ones most likely to be declined or lost in a dispute.