American Express has built a system that records what a customer told an AI shopping agent to buy, then checks the purchase against that record before payment goes through. The company calls it the Intent Intelligence Framework and hopes it becomes an industry standard, TechCrunch reported on 13 October.
Despite the name, what Amex describes is plumbing rather than a rulebook. It begins with a Create Intent API, which captures the customer’s instructions along with the limits attached to them, such as the merchant’s name, a maximum budget, and product preferences. Next, the customer proves who they are and approves the captured request. Amex issues an Intent ID, a unique reference for that request, plus a Proof of Intent Token that holds the broad limits the customer set.
The token is the part a reader can picture. The agent trades it for a secure digital payment credential, which it uses at checkout in place of the customer’s real card number. Amex’s own example: a customer asks for white Nike sneakers in size 9, and the agent tries to buy Adidas. The request no longer matches the signed intent, so Amex says it could stop the transaction. The catch happens before money moves, not weeks later in a dispute.
The same ID and token can follow the purchase through later stages, which Amex says could help with fraud prevention and dispute resolution. Hannah Bozian, Amex’s VP of Agentic Partnerships and Strategy, also pointed to a log of what was authorized, by which agent, and when.
Bozian argued the design matters most as agents get more freedom. Many shopping agents today ask for consent before buying, she said, but Amex expects some to act alone. Her example was reordering paper towels, where the only constraints are price and size. In that case a machine check stands in for the human click.
Now the limits of the news. The framework is one of the services in the ACE Developer Toolkit, which Amex launched in April. TechCrunch’s report names no launch date and no partner companies; it says only that Amex is “already in discussions” with firms elsewhere in payments. That is some distance from merchants and agent builders enforcing anything. Amex also says its approach differs from others because it pulls intent earlier into the transaction, which implies rival systems exist, though the piece does not name them.
The report is also silent on liability. It does not say who absorbs the loss when an agent buys the wrong thing, or what happens if the agent records the customer’s wishes badly and the customer signs off anyway. A check that blocks mismatched purchases shrinks the number of those arguments. It does not settle them.
Amex says it has not yet seen disputes rise because of agent shopping, so this is a build-ahead move, not a response to losses. Its supporting numbers come from its own Trendex survey, and both measure appetite for research, not for buying. Amex told TechCrunch that 66% of the U.S. consumers it polled say they are likely to hand an agent the job of finding a better price, and 61% a product suggestion, within a year. Neither figure shows people handing over a card.
The pattern to note is who is writing the rules. A payment network is defining how an agent proves it was allowed to spend, and the AI companies are not the ones setting the terms. Any team building an agent that pays should expect to be asked for something like this proof of intent, and should find out now who carries the loss when it is wrong.
Reported by TechCrunch (Sarah Perez) on 13 October 2026
