American Express launches framework to stop AI agents from making unauthorized purchases

AI agents are increasingly making payments on behalf of users, but the unpredictability of large language models creates a risk of unauthorized transactions and errors. American Express has tackled this challenge by launching the Intent Intelligence Framework — the first systematic approach to controlling buyer intent throughout the entire payment chain.

Intent

Intelligence Framework: a new standard for agentic commerce

The Intent Intelligence Framework is built into the ACE Developer Toolkit (Agentic Commerce Experiences), which American Express released in April. The framework aims to become an industry standard for AI payments and promises to make agentic transactions "as safe, reliable, and convenient as a regular card payment," according to Hannah Bozian, vice president of agent partnerships and strategy at American Express.

The company hasn't yet seen a rise in disputes stemming from agentic commerce, but decided to develop the technology proactively. The inspiration comes from Amex Trendex surveys: 66% of American consumers said they're willing to use AI agents to find the best price for a product or service. Another 61% are willing to use agents for recommendations within the next 12 months.

How the system works: three stages from intent to payment

The process begins with the Create Intent API — a tool that captures the buyer's instructions right at the start. The API collects the merchant name, maximum budget, product specifications, and other constraints. How intent is gathered depends on the AI model, but the system is designed to support any agentic commerce model.

The buyer then authenticates their identity and confirms that the agent correctly understood their intent. At the confirmation stage, the system generates an Intent ID — a unique intent identifier — and a Proof of Intent Token, a token containing the high-level constraints set by the user.

This token is used to obtain a secure digital payment credential, which the agent applies at checkout instead of a real credit card number. The payment stays protected and is guaranteed to match the user's authorization. The Intent ID and token are also used for fraud management and dispute resolution, providing a complete authorization log.

Real-world examples: from errors to autonomy

Amex's key innovation is intercepting intent early in the payment process, before the transaction is completed. This prevents errors. Bozian gave a concrete example: a user asks an agent to buy white Nike sneakers in size 9, but the agent mistakenly orders Adidas shoes instead. American Express's system would block such a transaction, since it doesn't match the authorized intent.

As agentic commerce evolves, even more autonomous scenarios are possible. A user might let an agent reorder everyday essentials — paper towels, toilet paper, and the like — independently, where the main parameters are price and size, and the merchant and brand aren't critical. In such cases, the system acts as a safeguard, verifying that the payment matches the original intent.

American Express is already in talks with partners across the payments ecosystem about adopting the ACE Developer Toolkit and agentic tokens. The company plans to continue actively developing the framework as the agentic commerce market evolves.