American Express Launches an "Intent Intelligence Framework": Before an AI Agent Buys for You, Budget, Merchant and Product Preferences Are Locked Into a Verifiable Token

On October 13 American Express launched its Intent Intelligence Framework to stop AI agents from making purchases users didn't authorize or that go against their intent, and hopes it becomes an industry standard for agent transactions. The flow: a Create Intent API records the user's purchase instructions and constraints, such as merchant, maximum budget and product preferences; after the user authenticates and approves, the system generates a unique Intent ID and a Proof of Intent Token carrying the main constraints; the agent then exchanges that token for a secure digital payment credential to check out, never touching the user's real card number. The Intent ID and token can also support later fraud prevention and dispute resolution. Amex's example: if an agent asked to buy white Nike sneakers in size 9 submits a request for Adidas shoes, the transaction can be blocked. The framework is now part of Amex's ACE developer toolkit launched in April; Amex says it is in discussions with payments ecosystem partners but named none.

One transaction, three parties

A traditional card transaction has two parties: consumer and merchant. Hannah Bozian, Amex's VP of agentic partnerships and strategy, told TechCrunch that transactions are about to have three: a consumer, a merchant and an agent. That raises the obvious question: when the agent buys the wrong thing or too much, whose problem is it? On October 13 American Express launched its Intent Intelligence Framework, built on the idea of locking down what the user actually wants before the agent acts.

How the flow works

  1. The agent calls the Create Intent API to record the user's instructions and constraints: which store, maximum spend, which products are preferred;
  2. The user authenticates and approves that "intent";
  3. The system generates a unique Intent ID and a Proof of Intent Token carrying the main constraints;
  4. The agent exchanges that token for a secure digital payment credential to check out, never getting the user's real card number;
  5. Later fraud checks and dispute resolution can trace back to the Intent ID and token.

Amex's example is concrete: if a user asks an agent for white Nike sneakers in size 9 and the agent submits an Adidas order, that transaction can be blocked outright. Bozian says what sets the approach apart from others is that it pulls intent earlier into the transaction flow.

What's still missing

The framework is now part of ACE (Agentic Commerce Experiences), the developer toolkit Amex launched in April. Amex says it is talking with partners across the payments ecosystem about adopting it, but named none and didn't spell out how liability is split when an agent gets a purchase wrong. Amex says it hasn't yet seen an increase in disputes tied to agent shopping.

For developers building shopping agents, this offers a ready-made "authorization credential" pattern to integrate with: user-approved constraints travel with the transaction, which is easier for users to trust than an agent roaming free with a full card number. For users, until payment providers actually support intent checks like this, it's wise to set your own per-transaction limit before giving an agent payment access.

via: TechCrunch report