American Express published an Intent Intelligence framework on October 13, 2026, describing how an AI shopping agent could carry a customer’s approved merchant, budget and preferences into a card transaction. The design is meant to give issuers and merchants more context than a bare payment credential. It does not mean every agent or merchant can use the full flow today.
What American Express published
The framework sits inside American Express’s Agentic Commerce Ecosystem, or ACE. In the company-issued release, American Express says a customer would set purchase instructions through a Create Intent API. Those instructions can include the merchant, a maximum budget and product preferences.
After the customer authenticates and signs off, the system creates two pieces of context:
- an Intent ID, which identifies the approved purchase request;
- a Proof of Intent Token, which lets other participants check that the customer authorized the instructions.
The company says the shopping agent can then request a digital payment credential that is limited to the approved intent. The agent does not need to receive the customer’s actual card number.
This is a framework and developer specification, not a report of a universal consumer rollout. American Express’s ACE page lists the Account Enablement, Intent Intelligence and Payment Credentials specifications as available, while other pieces remain under development.
How the proposed intent flow works
The sequence is designed to keep customer approval attached to the purchase as it moves from shopping to payment:
- The agent and customer settle on the merchant, product preferences and spending ceiling.
- The customer authenticates and approves those instructions.
- American Express creates an intent record and proof token.
- The agent requests a scoped digital payment credential.
- The intent information can travel into fraud checks, authorization and any later dispute.
A scoped credential matters because it can be tied to the approved task instead of acting like a reusable card number. For example, a customer could authorize an agent to buy one item from one merchant below a defined amount. A later attempt outside those terms would not match the original intent.
The approach overlaps with questions in our guide to agentic payments: who authorized the purchase, what limits applied, and what evidence exists if the transaction is challenged.
What is available now—and what is still planned
American Express’s public ACE page says the Account Enablement, Intent Intelligence and Payment Credentials specifications were available as of April 14, 2026. The October 13 publication explains the Intent Intelligence framework and its role in trusted agentic commerce; it should not be read as the first availability date for every underlying specification.
Two other ACE services remain marked as under development:
- Agent Registration, intended to identify registered agents and their capabilities;
- Cart Context, intended to give more detail about the merchant, goods or services, and purchase context.
American Express also describes a future Agent Purchase Protection program for eligible US consumer and small-business card purchases made through registered agents after authentication and confirmed intent. The company says coverage would be subject to terms, eligibility and claims review. It does not give shoppers permission to assume that protection applies to an agent purchase today.
Why issuers want proof of intent
A normal online card authorization can show the merchant, amount and device or network signals. An AI agent introduces another actor between the buyer and the checkout. That raises basic questions: Did the customer approve this exact merchant? Was the amount within the limit? Did the agent switch products? Was the agent allowed to complete the purchase or only research it?
American Express’s framework tries to answer those questions with a signed record of the customer’s instruction. The company says the intent controls can continue through fraud prevention, authorization and disputes rather than disappearing after checkout.
That could help separate three different failure modes:
- an agent acting outside the customer’s approved limits;
- a merchant charging more than the agent presented;
- a customer changing their mind after an otherwise authorized purchase.
The framework does not eliminate disputes. It aims to give each participant a clearer record of what was approved.
What developers and merchants should check
Teams evaluating agentic checkout should not treat a payment credential as the whole integration. They also need to understand:
- how the customer authenticates the intent;
- which fields are signed and which party can verify them;
- whether the credential is restricted by merchant, amount, time or purpose;
- how changed carts and substitutions are handled;
- what evidence is available to support a refund or dispute;
- which ACE services are available and which remain in development.
Our agentic checkout refunds and chargebacks guide covers the operational questions that remain after an agent presses Buy. The agentic commerce protocol tracker follows how the competing specifications fit together.
What changes for cardholders
For cardholders, the useful idea is not “AI shops for you” by itself. It is that the authorization can include boundaries the customer actually approved. Those boundaries may make it easier for an issuer or merchant to understand whether an agent followed instructions.
Customers still need to read the final merchant, amount and purchase terms before approving an intent. A future protection program may add another layer, but American Express says it is still being developed and will have eligibility and claims conditions.
