American Express Unveils an "Intent Intelligence Framework" to Lock In and Verify a Customer's Purchase Intent Before an AI Agent Pays

On October 13, American Express introduced a new framework to prevent AI agents from making unauthorized purchases. TechCrunch reports that personal AI agents increasingly make transactions on behalf of their users. But large language models are nondeterministic, so AI bots can do what they are not supposed to, which raises the chance of unauthorized transactions. To get ahead of this, Amex built the "Intent Intelligence Framework", which captures and authenticates a customer's purchase intent before an agent completes the transaction. The idea is that the customer's purchase instructions should be carried through the whole transaction lifecycle, from the purchase to everything that follows. Hannah Bozian, Amex's VP of Agentic Partnerships and Strategy, told TechCrunch that these new transactions have three parties, a consumer, a merchant and an agent, and that Amex wants them to stay "seamless, trusted, and secure, the way our card members expect them to be." Amex hopes the framework will become an industry standard.

The flow is specific. It begins with a Create Intent API that captures the customer's purchase instructions along with other constraints, such as the merchant's name, maximum budget and product preferences. How the intent is collected may vary with the AI model involved, but Amex plans to make the framework flexible enough for all agentic commerce models. The customer then authenticates their identity and signs off on the intent that the agent captured. After that an Intent ID, a unique identifier for the intent, is created, together with a Proof of Intent Token that holds the customer's high-level constraints. The token is used to obtain a secure digital payment credential, and the agent uses this credential at checkout instead of the customer's real card number. Amex says the purchase stays secure and stays aligned with what the customer authorized the agent to do. The ID and token can also serve other parts of the transaction process and help firms with fraud prevention and dispute resolution.

Bozian said Amex's approach differs from others because it pulls intent earlier into the transaction flow, which helps reduce some risk before it occurs. Her example: a customer asks an agent to buy white Nike sneakers in size 9, but the agent submits a request for shoes from Adidas. Amex could then stop the transaction because the agent made a mistake. She also noted that many agents today ask for customer consent before buying, but Amex sees a future where agents act more on their own in some cases. A consumer might let an agent reorder household staples such as paper towels, where the only real limits are price and size, not merchant or brand. In those cases it makes sense to have technology in the loop for an extra check that the purchase matches the customer's intent. A log of what was authorized, by which agent and when, could also help with fraud on the agent's behalf.

The framework extends Amex's April launch of Agentic Commerce Experiences, the ACE Developer Toolkit, and the intent framework is now one of its services. Amex says it has not yet seen an increase in disputes from agentic commerce, but wants to build the technology early as this commerce grows. Part of the reason is that consumers say they are open to it. In a recent Amex Trendex report, 66% of surveyed US consumers said they are likely to use AI agents to research the best price for a product or service, and 61% said they are likely to use one for a recommendation over the next 12 months. Amex is already in talks with partners across the payments ecosystem about implementing ACE, which includes the agentic tokens that contain Amex's payment credentials, and it plans to keep developing ACE as the ecosystem changes. Bozian discussed the framework in more detail on the Industry Stage at TechCrunch Disrupt 2026.

How should we read this? Its value is that it turns "what did the user agree to" into data that can be checked, instead of a vague hope that the agent will follow instructions. But there are limits. This is a proposal from one card issuer. To become a standard it needs merchants, other payment networks, and AI model and agent developers to take part. The report gives no rollout date or pilot size, and it does not explain what happens when an agent misreads an intent that the user did confirm. For consumers, the practical steps today are to let agents use payment methods with a spending cap, keep transaction alerts on, and stay careful when an agent asks you to link a real card number.