How Amex plans to keep AI purchases within a customer’s intent

American Express is adding an Intent Intelligence Framework to its AI commerce toolkit. It records a customer’s purchase instructions and constraints, then uses them to check whether an agent’s transaction matches what the customer approved.

WTF Index NEUTRAL
◄ Terminator 1 Idiocracy 1 ►

This routine commerce update aims to keep agent purchases within customer-approved limits, with only a mild concern about growing agent autonomy.

How Amex plans to keep AI purchases within a customer’s intent

As AI agents take on more shopping tasks, American Express is introducing a way to check that purchases stay within the customer’s instructions. Its Intent Intelligence Framework records what a customer has authorized before an agent completes a transaction.

Capturing instructions before checkout

The framework starts with a Create Intent API. It captures the customer’s purchase request and constraints, which may include a named merchant, a maximum budget, or product preferences. The way an agent collects those instructions may differ by AI model, but Amex says the framework is intended to support different models of agentic commerce.

After the agent records the request, the customer authenticates their identity and approves the captured intent. The system then creates an Intent ID, a unique reference for that intent, and a Proof of Intent Token containing the customer’s high-level constraints.

The token is used to obtain a secure digital payment credential for checkout. The agent uses that credential instead of the customer’s actual card number. Amex says this links the payment to the purchase the customer authorized while helping keep the transaction secure.

A check against agent mistakes

Amex’s case for recording intent early is that a mismatch can be caught before payment. Hannah Bozian, the company’s VP of Agentic Partnerships and Strategy, described a customer who asks an agent to buy white Nike sneakers in a size 9, while the agent tries to order shoes from Adidas. The framework could stop the purchase because it does not match the request.

The Intent ID and token may also be useful beyond checkout. Amex says they can support processes such as fraud prevention and dispute resolution by giving companies a record of what was authorized, by which agent, and when.

That record could matter if a transaction is questioned later, including in cases involving fraud on behalf of the agent. The framework is designed to carry the customer’s instructions through the transaction lifecycle, rather than treating checkout as the only point where approval matters.

Why Amex is preparing now

Amex says it has not seen an increase in disputes tied to agentic commerce. It is developing the framework proactively as this kind of shopping grows, and as consumers show interest in using AI agents for parts of the buying process.

In a recent Amex Trendex report, 66% of surveyed U.S. consumers said they were likely to use AI agents to research the best price for a product or service. Another 61% said they were likely to use an agent for a product or service recommendation over the next 12 months, the company told TechCrunch.

Today, agents may ask a customer to approve a purchase before placing it. Bozian also anticipates more autonomous uses. A customer might authorize an agent to reorder household staples such as paper towels or toilet paper, where the key limits could be price and size rather than a specific brand or merchant. In that kind of arrangement, a system check can help ensure a purchase fits the permission already given.

Part of a broader commerce toolkit

The framework is being added to Amex’s Agentic Commerce Experiences, or ACE Developer Toolkit, which the company launched in April. The toolkit includes services such as agentic tokens that contain Amex payment credentials, alongside other tools for developers.

Amex says it is in discussions with partners across the payments ecosystem about implementing ACE. It plans to keep developing the toolkit as agentic commerce evolves, and hopes the Intent Intelligence Framework can become an industry standard for AI-powered transactions.

The approach places customer intent near the beginning of the transaction flow, where it can inform later checks. Whether it can work consistently across different agents and commerce models will depend on how the framework is implemented as the ecosystem develops.