Amex's Intent Intelligence Framework verifies purchase intent before AI agents check out

American Express has unveiled a framework designed to stop AI shopping agents from completing purchases their owners never asked for. Called the Intent Intelligence Framework, it captures and authenticates a customer's purchase instructions before an agent is allowed to transact, and according to TechCrunch, Amex wants it to become an industry standard for AI-driven commerce.

The concern it addresses is specific to how large language models behave: they are not deterministic, so an agent acting on a user's behalf can drift from its instructions, including spending money in ways the customer never authorized.

How the framework works

The flow starts with a Create Intent API, which records the customer's instructions along with any constraints attached to the request, such as a specific merchant, a maximum budget, or product preferences. How that intent is gathered may differ from one AI model to the next, but Amex designed the framework to be flexible enough to support any agentic commerce model, Hannah Bozian, the company's VP of Agentic Partnerships and Strategy, told TechCrunch.

The customer then verifies their identity and signs off on the intent the agent captured. At that point Amex issues an Intent ID, a unique identifier for the request, plus a Proof of Intent Token encoding the customer's high-level constraints. That token is exchanged for a secure digital payment credential, which the agent presents at checkout instead of the customer's real card number.

Because the token carries the customer's constraints, the transaction can be checked against them. Bozian gave the example of a customer asking for white Nike sneakers in a size 9 while the agent tries to buy Adidas shoes instead; in that case Amex could block the purchase because it does not match what was authorized. She said the approach differs from other implementations by pulling intent earlier into the transaction flow, which addresses risk before it materializes rather than after.

The Intent ID and token also have uses beyond checkout. Amex says they can support fraud prevention and dispute resolution, since there is a record of what was authorized, by which agent, and when — something Bozian noted could help address fraud carried out through agents.

Part of a broader toolkit

The framework is not a standalone product. It extends Amex's Agentic Commerce Experiences, or ACE Developer Toolkit, which the company launched in April. Amex told TechCrunch it has not yet seen a rise in disputes linked to agentic commerce, but is building the technology ahead of expected growth, pointing to its own Trendex survey in which 66% of U.S. consumers said they are likely to use AI agents to research the best price for a product or service, and 61% said they are likely to use one for product or service recommendations over the next 12 months.

Bozian also sketched a future where agents act more autonomously, such as re-ordering household staples like paper towels, where the real constraints are price and size rather than brand or merchant. In those cases, she argued, having technology in the loop to confirm each transaction matches the customer's intent becomes more important, not less.

Amex says it is already in discussions with partners across the payments ecosystem about implementing ACE, including the agentic tokens that hold its payment credentials, and plans to keep developing the toolkit as agentic commerce evolves. Bozian is scheduled to discuss the framework in more detail at TechCrunch Disrupt 2026's Industry Stage on October 13.

Why it matters

Agentic commerce is arriving faster than the guardrails around it, and payment networks sit at a natural chokepoint where intent can be verified. Amex's design offers a transferable pattern for the industry: turn human intent into a signed, machine-checkable artifact, give the agent scoped credentials instead of a raw card number, and enforce the customer's constraints at transaction time. Developers building shopping agents get a concrete interface to build against rather than improvised checkout workflows, and consumers get an audit trail when something goes wrong. Whether the Intent Intelligence Framework becomes a standard or one implementation among many, the underlying idea — that authorization must travel with the transaction — is likely to shape how AI agents are allowed to spend money.