AI agents making purchases on your behalf sounds convenient right up until one of them buys the wrong thing with your credit card. American Express is betting that problem is coming fast, and it wants to own the solution before the disputes start piling up.
As TechCrunch reported, Amex has introduced what it calls the “Intent Intelligence Framework,” a system designed to lock in a customer’s purchase instructions before an AI agent actually completes a transaction. The company wants this to become an industry standard. That’s an ambitious goal, and worth scrutinizing.
Here’s how it works. When a customer tells an AI agent to buy something, a Create Intent API captures that request along with any constraints, things like a specific merchant, a maximum price, or a preferred product. The customer then authenticates their identity and confirms the captured intent. From that point, the system generates two things: an Intent ID, which is a unique identifier tied to that specific authorization, and a Proof of Intent Token. That token is used to get a secure digital payment credential, meaning the AI agent never touches your actual card number. It uses a stand-in credential instead.
The token and ID travel through the transaction, which Amex says helps with fraud detection and dispute resolution down the line. If an agent tries to do something outside what was authorized, the system can flag or block it before money moves. Amex VP of Agentic Partnerships and Strategy Hannah Bozian gave a clear example: a customer asks for white Nike sneakers in a size 9, but the agent tries to order Adidas. Under this framework, Amex could stop that transaction because the agent deviated from what the customer actually approved.
So from a privacy and security standpoint, there are things here worth taking seriously. Using a tokenized credential instead of a real card number is a sound practice. It limits exposure. And the idea of capturing intent at the start of a transaction, rather than doing a post-purchase review, does address a real weakness in how agentic commerce is being built right now. Most systems just ask for a thumbs-up before buying. Amex is arguing that’s not enough.
But some questions remain open. Who controls the intent data once it’s captured? How long is it retained? What happens when an AI model misinterprets a customer’s instructions before they even reach the authentication step? Amex says it hasn’t seen a rise in disputes from AI-driven purchases yet, but the framework’s release is clearly preemptive. The company cited its own Trendex survey showing that 66% of U.S. consumers said they’d likely use AI agents to research prices, and 61% said they’d use one for product recommendations in the next 12 months. Those numbers signal real adoption is coming.
The framework is part of Amex’s broader ACE Developer Toolkit, which the company launched in April. That toolkit includes agentic tokens carrying Amex payment credentials and other tools for developers building commerce experiences on top of AI agents. Amex says it’s already in talks with partners across the payments ecosystem to implement ACE.
The bigger picture here is that the financial industry is racing to set norms for agentic commerce before regulators do. Whoever defines the standard for how AI purchase intent is captured and verified will have significant influence over how billions of transactions get processed. American Express clearly wants that seat at the table. Whether their framework actually protects users or just protects the network’s liability exposure is a question worth keeping in mind as this rolls out.