American Express is introducing Intent Intelligence, a framework designed to check what a customer authorized before an AI agent completes a purchase. The commercial problem is practical: a payment can be securely authenticated while still buying the wrong item.
The framework aims to carry the customer's instructions through the transaction rather than rely solely on an agent's interpretation. That could make automated purchases easier to trust, although American Express has not disclosed transaction volumes or a measurable financial benefit from the framework.
How purchase instructions become a payment control
TechCrunch reported that the process begins with a Create Intent interface, which records instructions such as a merchant, maximum budget and product preferences. The customer confirms the instructions and authenticates their identity.
The system then creates an Intent ID and a Proof of Intent Token. The token carries the high-level limits and supports obtaining a digital payment credential, allowing checkout without exposing the actual card number.
The distinction between identity and intent matters. Authenticating a person establishes who authorized activity. Recording that person's limits gives the issuer information about what the activity should be. Neither an authenticated user nor a valid card credential alone proves that the purchased product matches the request.
American Express's published technical explanation describes services for creating, updating and canceling authenticated intent. This makes permissions something that can change over the shopping process rather than a one-time instruction left inside a chatbot conversation.
The value proposition extends beyond fraud
In TechCrunch's interview, Amex executive Hannah Bozian described an agent instructed to buy one brand of sneakers but requesting another. It is an illustrative error, not a reported customer incident. Constraints can support a check before payment, while a preserved record could help resolve a later dispute.
The same logic applies to repeat household purchases. A customer might authorize an agent to reorder supplies within a price and size limit. The usefulness of those boundaries depends on reliable interpretation and enforcement by the participating systems.
Amex said it had not yet seen increased disputes from agentic commerce. The announcement is therefore a preventive design for a developing market, not remediation of a documented surge in losses.
Interest is not adoption
TechCrunch cited an Amex survey in which 66% of surveyed U.S. consumers said they were likely to use AI agents for price research and 61% for recommendations within the next year. Research and recommendations do not necessarily include permission to pay, and stated intentions are not completed transactions.
Amex says it is discussing implementation with payments partners and plans further development of its ACE toolkit, introduced in April. The next evidence should be named integrations, the portion of purchases using authenticated intent and outcomes when an agent violates a limit.
The framework addresses a real gap between secure payment and correct purchase. Its investment significance will depend on whether merchants and agents implement the controls at scale, without adding enough friction to discourage use.