American Express adds intent controls for AI agent payments
Amex introduces an intent framework for agent-led purchases
American Express has introduced the Intent Intelligence Framework, a system intended to prevent AI agents from completing purchases that fall outside a card member’s approved instructions. The framework has been added to American Express Agentic Commerce Experiences, or the ACE Developer Toolkit, which the company launched in April.
The initiative addresses a basic problem in agentic commerce: large language models are nondeterministic, so an agent handling a transaction may not always follow a user’s request precisely. American Express wants a customer’s purchase instructions to remain attached to the transaction from the initial request through checkout and later processes.
Hannah Bozian, vice president of Agentic Partnerships and Strategy at American Express, said agentic transactions involve a consumer, a merchant and an agent. The company aims to keep those transactions seamless, trusted and secure while creating controls for the agent’s role.
How the purchase-intent controls work
The process starts with a Create Intent API. It captures a customer’s purchase request and any constraints, including a merchant name, maximum budget and product preferences. American Express said the mechanism for collecting intent may vary by AI model, while the framework is intended to support different agentic-commerce models.
After the agent captures the request, the customer authenticates their identity and approves the recorded intent. The system then creates an Intent ID, a unique identifier for that authorization, 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 instead of providing the customer’s actual card number to the agent. American Express said the Intent ID and token can also be used elsewhere in the transaction lifecycle, including fraud prevention and dispute resolution.
Controls before checkout
American Express says its approach brings intent into the transaction flow earlier, with the aim of mitigating risk before a payment is made. Bozian offered an example in which a customer asks an agent to buy white Nike sneakers in size 9, but the agent submits a request for Adidas shoes. The framework could stop that transaction because it does not match the approved request.
The company has not seen an increase in disputes caused by agentic commerce, but is building the technology in anticipation of broader adoption. In an American Express Trendex survey, 66% of surveyed US consumers said they were likely to use AI agents to research the best price for a product or service, while 61% said they were likely to use them for recommendations during the next 12 months.
American Express also anticipates more autonomous use cases, such as reordering household staples where price and size may matter more than a particular merchant or brand. A record of which agent was authorized, for which task and at what time could help address agent-related fraud, Bozian said.
What businesses should prepare for
American Express is discussing ACE implementation with partners across the payments ecosystem and plans to develop the toolkit as agentic commerce evolves. For businesses deploying purchasing agents, the practical implication is to define approval limits, merchant and product constraints, authentication steps, and records for exceptions before permitting agents to transact on a customer’s behalf.