Amex builds a framework to vet AI agents' purchases before checkout

Personal AI agents are increasingly handling transactions on their owners’ behalf, but because large language models are nondeterministic, they can act in ways users didn’t intend — including making purchases they never authorized. American Express says its new framework is meant to get ahead of that problem by ensuring a customer’s original instructions carry through the entire transaction, from purchase to settlement.

Dubbed the “Intent Intelligence Framework,” the system is designed so that agentic transactions involve a consumer, a merchant, and an agent, while still remaining “seamless, trusted, and secure,” said Hannah Bozian, Amex’s VP of Agentic Partnerships and Strategy, in an interview with TechCrunch. The release builds on the company’s April launch of its Agentic Commerce Experiences (ACE) Developer Toolkit, a set of services that now includes the intent framework.

Amex said it has not yet seen a rise in disputes tied to agentic commerce, but wanted to build the technology proactively as that kind of commerce scales — consumers have signaled openness to it: in a recent Amex Trendex report, 66% of surveyed U.S. consumers said they are likely to use AI agents to research the best price for a product or service over the next 12 months, and 61% said the same for getting a product or service recommendation.

The framework starts with a Create Intent API that captures a customer’s purchase instructions and any constraints, such as merchant name, maximum budget, or product preferences; the exact collection method varies by AI model, but the system is meant to be flexible enough to support any agentic-commerce model, Bozian said. The customer then authenticates their identity and signs off on the captured intent, generating an Intent ID and a Proof of Intent Token carrying the customer’s high-level constraints. That token is used to obtain a secure digital payment credential the agent uses at checkout instead of the customer’s actual card number, and the Intent ID and token can also support fraud prevention and dispute resolution.

Bozian said the framework pulls intent earlier into the transaction than other approaches, which helps mitigate risk before it occurs — for instance, if a customer asks an agent to buy white Nike sneakers in size 9 but the agent requests Adidas shoes instead, Amex could block the transaction as a mismatch with the authorized intent. She also said a log of what each agent was authorized to do, and when, could help address agent-related fraud, and that Amex expects agents to eventually act more autonomously in lower-stakes, recurring categories such as reordering household staples. Amex is already in discussions with partners across the payments ecosystem to implement ACE, and said it plans to keep developing the toolkit as the agentic-commerce ecosystem evolves.

Why it matters

As AI agents take on more purchasing decisions, a standard for proving what a customer actually authorized could shape how disputes and fraud liability are handled across the payments industry, not just at Amex.