American Express Introduces Intent Intelligence to Prevent Unauthorised AI Agent Purchases

American Express has introduced a new security framework designed to prevent AI agents from making purchases that exceed or contradict a customer’s instructions. Called Intent Intelligence, the system records and verifies a customer’s authorisation before an AI-powered assistant completes a transaction.

The framework addresses a potential weakness in agentic commerce, where AI assistants can research products, select items, and make payments on users’ behalf. Because large language models do not always follow instructions consistently, an agent could purchase the wrong product, choose an unauthorised merchant, or exceed a customer’s spending limit.

American Express wants to establish a method to preserve customer authorisation throughout the payment process. The company also hopes its framework will become an industry standard as AI agents take on more purchasing responsibilities.

How American Express Intent Intelligence Works

The framework begins with a Create Intent API that captures the customer’s purchasing instructions and any attached restrictions. These can include a preferred merchant, a maximum spending amount, product specifications, or other conditions the AI agent must follow.

Customers then authenticate their identity and approve the recorded instructions. Once authorisation is complete, the system generates an Intent ID that uniquely identifies the request, along with a Proof of Intent Token containing the customer’s high-level purchasing constraints.

The token provides a secure digital payment credential for the AI agent to use at checkout. Instead of handling the customer’s actual credit card number, the agent completes the payment using that credential, while the transaction can be checked against the customer’s approved instructions.

This adds a verification step beyond simply asking an AI assistant to confirm a purchase. American Express says the framework brings customer intent into the transaction process earlier, allowing certain errors to be detected before payment is completed.

Hannah Bozian, American Express vice president of Agentic Partnerships and Strategy, described a scenario in which a customer asks an AI agent to purchase white Nike sneakers in size 9, but the agent selects Adidas shoes instead. Under the proposed framework, the mismatch could be identified, and the transaction prevented because it doesn’t satisfy the original authorisation.

AI Shopping Agents Could Receive More Purchasing Freedom

American Express expects AI agents to become more autonomous as consumers grow comfortable delegating routine purchases. Instead of approving each transaction individually, customers might authorise an assistant to reorder household essentials within specified limits.

For example, a customer could let an AI agent buy paper towels or toilet paper as needed, as long as the items meet approved size and price requirements. In those cases, the framework would provide an additional check against the original instructions without requiring the customer to supervise every purchasing decision.

The system also creates records linking purchase authorisation to a specific agent and transaction. American Express says those records could support fraud prevention, transaction investigations and dispute resolution when questions arise about what the customer actually permitted.

The company said it has not observed an increase in disputes attributable to agentic commerce. The company developed the framework to address potential risks before AI-powered purchasing becomes more widespread.

Amex Expands Its Agentic Commerce Developer Toolkit

Intent Intelligence becomes part of American Express’s Agentic Commerce Experiences (ACE) Developer Toolkit, introduced in April 2026. The toolkit includes services designed to support AI-driven payments, including agentic payment tokens that enable secure transactions using American Express credentials.

The company is already discussing ACE implementation with partners across the payments industry. According to Bozian, the intent framework is being designed to accommodate different AI systems and shopping experiences rather than depend on one particular agent or model.

Consumer interest provides some context for the initiative. In an American Express Trendex survey, 66% of U.S. respondents said they were likely to use AI agents to research the best prices for products or services. In comparison, 61% said they were interested in receiving AI-generated product or service recommendations over the next 12 months.

Those findings concern shopping research and recommendations rather than completed purchases, but they indicate consumer willingness to involve AI assistants in purchasing decisions. American Express is positioning its new framework to address the additional security requirements that arise when those assistants move from recommending products to paying for them.

The company plans to continue developing ACE as agentic commerce evolves. Its immediate focus is establishing a consistent way to authenticate customer instructions, verify transactions against those instructions and preserve authorisation records throughout the payment lifecycle.