American Express Opens Framework for Verifying AI Agent Purchases

By James Calloway

American Express published its Intent Intelligence framework on Tuesday, Oct. 13, giving card issuers a way to verify purchases made by AI-powered shopping agents. The framework has no live integrations yet, the company confirmed to Digital Commerce 360.

Amex released the technical documentation so other payment and technology providers can adopt the approach, not to launch new capabilities. The company said its goal is to "help advance a common industry approach for intent in agentic commerce."

"Providing authorized purchase intent before an agent transacts can give issuers greater confidence that an agent is acting as a customer intended, while also helping issuers proactively manage risk and reduce the chance of costly disputes across the payments ecosystem," Luke Gebb, head of global innovation at American Express, said in the announcement.

Why does intent matter for AI purchases?

The framework records a shopper's purchase instructions before an agent buys anything. Card issuers can use those records to detect fraud, approve payments and resolve disputes. Retailers can use them to determine whether an ordering mistake originated with an AI agent.

That distinction matters to consumers. In an August survey of 1,000 U.S. adults by payments company NMI, 58% worried an AI shopping agent could buy the wrong item, size, color or brand. Another 60% feared an agent could overspend.

Amex first introduced Intent Intelligence in April as one of five services in its Agentic Commerce Experiences (ACE) developer kit. The company said the framework "differs from current implementations" by sharing a customer's authorized purchase intent to the issuer "earlier in the transaction lifecycle — before an agent executes a purchase."

How does the framework work?

American Express broke the process into four steps:

1. Create and confirm intent. Amex's Create Intent API captures what the shopper wants to buy, where the agent can shop and how much it can spend. The shopper verifies identity and approves the instructions.
2. Establish proof of intent. The system generates an Intent ID identifying the approved instructions and a Proof of Intent Token representing key purchase limits.
3. Scope the payment credential. The agent uses the token to obtain a secure digital credential tied to the instructions, then uses it at checkout instead of the shopper's card number.
4. Apply intent throughout the transaction. The ID and token check whether a purchase request matches the authorization, supporting "fraud prevention, authorization decisioning and dispute resolution."

Shoppers can set maximum spending amounts, expiration dates, merchant preferences and timing conditions — such as buying when a product returns to stock or on Cyber Monday. The framework also handles complex purchases: an agent could buy flight insurance only if it secures the tickets, or complete multi-merchant transactions such as shoes from one retailer and socks from another. The system works across agentic commerce models regardless of who collects the instructions — an agent, a wallet provider or another intermediary can pass them to the issuer.

What about purchase protection?

Amex's April ACE announcement outlined "Agent Purchase Protection" for eligible transactions from registered AI agents that share authenticated purchase instructions, aiming to reduce merchant chargebacks and disputes. That protection is not yet available. Amex confirmed it has no ACE integrations using Intent Intelligence, which transactions need to qualify. Card members retain existing protections on their eligible Amex cards.

Blame assignment is already shifting. In NMI's August survey, 62% of consumers said they would blame the AI provider if a tool purchased before they were ready, 34% would blame the retailer and 24% would blame the payment provider.

Where does this fit in agentic commerce?

American Express joins a crowded field. Visa launched its Intelligent Commerce program in April 2025, working with OpenAI, Microsoft, Anthropic, Stripe and Samsung. Mastercard introduced Agent Pay the same month, supporting consumer and B2B purchases through conversational AI platforms.

Technology companies are building parallel standards. OpenAI and Stripe co-developed the Agentic Commerce Protocol (ACP). Google's Agent Payments Protocol (AP2) uses digitally signed authorization records, with Salesforce, Shopify, Cloudflare and Etsy among early adopters. Google followed in January with its Universal Commerce Protocol, developed with Shopify and other partners; Newegg integrated it on Oct. 12, joining Wayfair, Walmart and Macy's in supporting UCP checkout through Google's AI Mode and Gemini.

Until Intent Intelligence gains live integrations, the framework functions as an open invitation: Amex is betting that standardized purchase-intent records become the industry's common defense against agent-driven disputes.