- American Express introduced the Intent Intelligence Framework, which records and verifies a customer's purchase instructions before an AI agent completes a payment.
- It matters because language-model agents can misfire, and a card network can only catch that if it knows what the customer approved.
- Leaders should ask payment partners how agent purchases are authorized, logged and disputed, and who owns the record of intent.
Card checkout has long asked one question: is this really the cardholder? AI shopping agents add a second one. Is this what the cardholder asked for? That shift, from proving identity to proving intent, is the idea behind a framework American Express announced on October 13.
What Amex announced
Amex introduced the Intent Intelligence Framework. It is designed to capture and authenticate a customer's purchase intent before an AI agent completes a transaction. Amex hopes it becomes an industry standard for AI-powered transactions.
The framework is now one of the services in the ACE Developer Toolkit, which Amex launched in April. TechCrunch, which reported the news and interviewed Amex, notes that language models do not produce the same output every time. That means an agent can misfire. In shopping, that could mean an unauthorized purchase.
Hannah Bozian, Amex's VP of Agentic Partnerships and Strategy, told TechCrunch that these transactions will have three parties: a consumer, a merchant and an agent. She said they should still be "seamless, trusted, and secure."
How it works, step by step
The process starts with a Create Intent API. An API is a standard way for software systems to talk to each other. This one captures the customer's instructions and any limits they set, such as the merchant's name, a maximum budget and product preferences.
Next, the customer authenticates and signs off on the intent the agent captured. The framework then creates two things. One is an Intent ID, a unique label for that request. The other is a Proof of Intent Token, which holds the customer's high-level constraints.
The agent then uses that token to get a substitute payment credential. At checkout, the agent pays with the substitute, and the customer's real card number stays out of the exchange. Amex says this keeps the purchase secure and in line with what the customer approved.
Bozian gave an example of a mismatch. A shopper wants white Nike sneakers in size 9. The agent files an order for Adidas shoes instead. Amex says it could block that sale, because the agent made a mistake.
The idea: a purchase order for machines
Finance teams know this pattern. A company issues a purchase order before it buys, so the invoice can be matched against what was approved. Amex's framework works in a similar way for consumers and their agents. The approval is written down first, and every later step is checked against it.
This shows where the risk now sits. Once an agent acts for you, the dispute is no longer only about whether the card was stolen. It is about whether the agent did what you said. A card network can only judge that if a record of your instruction exists.
Amex says the Intent ID and token can be used at several points in a transaction, including fraud prevention and dispute resolution. Bozian added that a record of which agent received approval, and when, could also help with fraud involving agents.
Why the check matters more as agents act alone
Bozian noted that many agents today ask for consent before buying. Amex foresees agents that may act much more independently in some cases. Her example was re-ordering household staples such as paper towels, where the only real constraints are price and size.
In those cases no person reads each order. Bozian argued that software should run an added check that the order fits what the customer wanted.
These are Amex's own survey figures, reported by TechCrunch. They measure stated likelihood among surveyed U.S. consumers, not actual behavior.
What is still unknown
The framework has been released as part of Amex's ACE toolkit. Rollout is a separate matter. Amex told TechCrunch it is already in talks with partners across the payments ecosystem about implementing ACE, including the agentic tokens that carry its payment credentials. No partner is named, so adoption beyond Amex is unconfirmed. Amex says it will keep developing ACE as agentic commerce evolves.
Amex also told TechCrunch that agent-driven commerce has not yet pushed its dispute numbers up. It built the framework ahead of that risk. It is not a response to a measured wave of losses.
TechCrunch's report does not say how a vague instruction would be recorded. It also does not say who else would hold the intent record, or how merchants and other card networks would adopt it. Bozian said the way intent is collected may vary by AI model, and that the framework aims to support all agentic commerce models. Amex also claims its design sets it apart by capturing intent at an earlier stage of the sale. That is the company's claim.
Questions to put to your team
If your company sells online, ask your payments provider how it will treat purchases made by an agent. Ask whether it can see what the customer authorized, and what evidence you would have in a dispute.
If your company builds or deploys agents, ask where the customer's instruction is stored. Ask whether a record exists that is separate from the agent's own account of events.
For everyone else, ask who carries the loss when an agent buys the wrong thing. TechCrunch's report does not settle that question.
A signature once proved that a person agreed. In agent commerce, the lasting question may be whether anyone wrote down what they agreed to.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: TechCrunch.
