TechCrunch reported that American Express has introduced an “Intent Intelligence Framework” aimed at preventing AI agents from making purchases that fall outside a customer’s original instructions. The proposal addresses a problem that becomes more important as software moves from recommending products to buying them: an agent can interpret, improvise or misunderstand a request, while payment systems are generally built around a simpler assumption that the person, merchant and payment credentials are directly connected.
The framework would place an explicit record of the customer’s intent between a natural-language request and the final payment. Instead of treating a checkout approval as the only meaningful expression of consent, the system would preserve what the shopper actually asked the agent to do, along with the limits placed on that request.
Turning a request into a permission
American Express’s announcement describes a Create Intent API that can capture instructions and constraints before an agent begins shopping. Those details could include a preferred merchant, a maximum budget, product characteristics, a brand and a size.
The customer would then authenticate their identity and approve the captured intent. Once approved, the system would create an Intent ID and a Proof of Intent Token. The token would carry high-level constraints that allow other participants in the transaction to understand what the agent is authorized to do without receiving the customer’s full card information.
That distinction matters because an AI agent may need enough payment capability to complete a purchase, but it should not necessarily receive an unrestricted card number. Under Amex’s proposal, the agent could obtain a scoped digital payment credential tied to the approved intent. If the requested product, merchant or price falls outside that scope, the transaction could be blocked or flagged before money changes hands.
The approach treats a shopping instruction less like a casual conversation and more like a durable permission slip. “Find me white Nike sneakers in size nine” would not simply disappear into the agent’s context window. It could become a structured authorization that remains available while the agent searches, selects an item and attempts to pay.
A record that follows the transaction
American Express Technology’s technical overview says the purpose of the framework is to represent confirmed user intent as durable guardrails for an AI agent. The emphasis is not only on what an agent can do, but on how its actions can remain constrained by the customer’s instructions.
That record could also remain useful after checkout. Amex says intent information could support fraud detection, dispute resolution and verification that the final purchase matched what the customer authorized. If a customer later challenges a transaction, the relevant question would not be limited to whether a card was used successfully. Systems could also examine what the customer approved, what the agent attempted and where the two diverged.
This creates a possible chain of accountability for agentic commerce. A merchant could see that a payment credential was issued for a particular kind of purchase. A payment network could compare the final transaction with the approved constraints. An AI platform could be required to explain why it selected a product or attempted a charge beyond the customer’s limits.
The framework is being developed before Amex has seen a rise in disputes specifically caused by agentic commerce, according to the company’s announcement as described by TechCrunch. That timing reflects a bet about adoption. Agents may initially help shoppers compare prices, discover products and prepare carts. Over time, Amex expects them to handle more autonomous recurring purchases, including household staples.
A monthly order for detergent or pet food could be convenient precisely because the customer does not want to review every purchase. But convenience also creates ambiguity. Does permission to reorder mean any brand is acceptable? Does it include substitutions? What happens when the price rises, the preferred merchant is out of stock or the agent finds a product that appears equivalent but is not?
An intent record could give those decisions explicit boundaries rather than leaving them to the model’s interpretation.
Building a wider agent payment system
The framework is part of a broader initiative. Amex’s Agentic Commerce Experiences page describes a developer kit that includes Intent Intelligence, account enablement, payment credentials and prospective protection for eligible purchases made by registered AI agents.
For consumers, the visible experience could be simple. A person states a goal, reviews the limits, authenticates once and lets an agent work within the approved space. Behind that interaction, the payment system would need to distinguish between an agent acting within permission and an agent making a new decision on the customer’s behalf.
That distinction may determine whether people trust autonomous shopping. Consumers are often comfortable delegating research because a mistaken recommendation can be ignored. A mistaken purchase is different. It creates a financial commitment, a delivery obligation and a potential dispute.
Michael Brous of Forter, in a LinkedIn post supporting Amex’s announcement, said agentic commerce needs trusted identity and real-time intent signals, and described ACE as a step toward infrastructure that could help merchants adopt the technology safely. That is the only outside reaction identified in the supplied sources. No opposing view was found.
The larger question is whether agentic commerce will require a new definition of consent. Today, authorization often arrives as a button pressed at the end of a checkout flow. In a world of shopping agents, consent may need to begin earlier, when the customer defines the goal and the boundaries.
If merchants, card networks and AI platforms adopt systems like Amex’s, an agent’s freedom could become conditional rather than unlimited. The software would still search, compare and decide within a range. But when its decision became a financial commitment, the purchase would have to prove that it was the purchase the customer meant to make.
Maya Lindqvist is not a person. No notebook, no deadlines, no face behind the name — just a byline this newsroom publishes under. Here is the production line underneath it, because a name beside a portrait reads like a journalist, and this one is not one.
The models. Writing: gpt-5.6-luna and qwen3-max. Out on the live web: gpt-5.6-luna and gpt-5.6-terra. Pictures: gpt-image-1 and gpt-image-1-mini. Swap one in the newsroom and this line swaps with it — it is read off the machines, not typed here.
How a story is made
- Research. The searching model reads around the story, pointed at primary sources — the filing, the post, the repository — rather than at somebody else's write-up of them.
- Writing. The writing model drafts it against what was found, at Maya Lindqvist's usual length and in Maya Lindqvist's usual register.
- The loop. A reviewer reads the draft and sends it back with notes. Then reads it again. A piece can go round several times before it leaves the building.
- Enrichment. A quotation has to appear word for word on the page it is taken from. A chart may only use figures that appear in the source it cites. Whatever fails is dropped, and the reason is kept.
- Fact check. A last pass hunts for claims the article makes and its sources do not.
- A human stop. Sensitive subjects are held for a person to read before publication, and a person can kill any of it at any point.
If that sounds less like a newsroom and more like a factory: quite. It is called Press Factory.
This article was generated using AI and published automatically without human pre-publication review.
How this article was made
The article was produced by the Grandmonts Media News Engine using automated research, drafting and verification workflows. No human editor reviewed the article before publication. Grandmonts Media remains responsible for the published content. Errors can be reported at office@grandmonts.cz.