American Express on October 13, 2026 introduced a new mechanism designed to stop AI agents from making unauthorized transactions on behalf of shoppers — the Intent Intelligence Framework. The system captures and confirms the customer's purchase intent before the transaction completes; Amex expects the approach to become an industry standard for AI-driven transactions (TechCrunch).

Personal AI agents are increasingly taking over payments on behalf of users, but the probabilistic nature of large language models can lead them to act unexpectedly. Amex is trying to address this problem in advance: the customer's shopping instructions are preserved throughout the entire transaction flow — from payment to the steps that follow.

Intent Is Captured at the Start of the Transaction

The framework begins with the Create Intent API. This API records the customer's shopping instructions together with constraints such as the merchant name, maximum budget, and product preferences. The customer then verifies their identity and approves the recorded intent.

Next, the system creates an Intent ID — a unique intent identifier — and a Proof of Intent Token. The token carries the customer's high-level constraints and is used by the agent to obtain a secure digital payment credential. At checkout, the agent uses that credential instead of the customer's real card number. As a result, the purchase stays secure while remaining strictly within the scope the customer authorized for the agent.

The Intent ID and token can also be used in other parts of the transaction: according to Amex, they help companies prevent fraud and resolve disputes.

What Sets It Apart: Intent Is Asked for Earlier

Amex says its approach differs from competitors' by introducing intent into the transaction flow earlier — which makes it possible to reduce risk before it materializes.

"The company wants agentic transactions to be smooth, trustworthy, and secure — just as our cardholders expect," said Hannah Bozian, Amex's vice president of agentic partnerships and strategy.

Amex gave an example: if an agent is asked to buy size 9 white Nike sneakers but sends a request for Adidas shoes, Amex will block the transaction because of the agent's mistake.

Bozian added that today many agents ask for the customer's consent before purchasing, but in the future they may act far more independently. For example, a consumer could allow an agent to reorder everyday essentials such as paper towels or toilet paper — categories where the only constraints would be price and quantity, not merchant or brand. In her view, it is appropriate to have technology in the chain that verifies transactions against customer intent in such cases.

Additionally, keeping a log of which agent confirmed which action and when can help fight fraud committed on behalf of agents.

Five Services in the ACE Suite

The Intent Intelligence Framework builds on Amex's Agentic Commerce Experiences (ACE) developer suite, launched in April. According to VentureBeat, ACE gives developers five integrated services: agent registration, account linking, intent intelligence, payment credentials, and cart context.

Once an agent is registered, the system links the user's Amex account to the agent and authorizes it to make purchases. At the intent-intelligence stage, Amex forms a document it calls an "intent contract": the user defines what the agent will do, then the ACE system creates the Intent ID and Proof of Intent Token — proof of authorization should a dispute arise.

At the payment stage, once the agent finds the product the customer requested, it requests payment credentials. These are issued as a single-use token tied to the customer's constraints: for example, if the customer only wants to spend $500, the token will not authorize a $600 purchase.

Amex is already in talks with partners across the payments ecosystem about rolling out ACE. The company plans to actively expand the suite as the agentic-commerce ecosystem grows.

Consumers Are Open to Agentic Commerce

According to Amex's Trendex survey, 66% of U.S. consumers said they could use AI agents in the next 12 months to find the best price on a product or service, and 61% said they could turn to an agent for a recommendation.

The company has not yet seen a rise in disputes from agentic commerce, but it is building the technology ahead of time as this commerce scales. The topic of agentic shopping assistants has appeared in the news before — for instance, we wrote about TikTok's AI shopping assistant with one-click checkout.