New Amex tool gives the AI agent a token instead of your card number
American Express is launching the Intent Intelligence Framework on October 13, 2026, which captures and authenticates what the customer actually asked the agent to buy — and replaces the card number with a secure digital payment credential tied to that intent. In an agentic commerce market where users readily let AI find prices but barely dare let it pay, the card network is trying to move trust checks from after the fact to before it.
American Express (Amex) is introducing a new framework designed to capture and authenticate the customer's purchase intent before an AI agent completes a transaction, TechCrunch reports ([1]). The framework is called the Intent Intelligence Framework and is part of the ACE Developer Toolkit (Agentic Commerce Experiences), which Amex launched in April 2026. Amex hopes the framework will become an industry standard for AI-driven transactions — a hope, not an achieved status.
The launch coincided with TechCrunch Disrupt 2026, where Amex's vice president of agentic partnerships and strategy, Hannah Bozian, was set to discuss the framework. According to TechCrunch, she explained that agentic transactions have three parties: "These new transactions will have a consumer, a merchant, and an agent" ([1]).
The problem: wrong orders and reluctance to hand over the card
The backdrop is an increasingly visible trust gap in agentic commerce. Business Insider has spoken with shoppers who use AI agents like Instinct and Muse: some save time and find deals they would not have found themselves, but even the impressed users employ the agents cautiously. Sometimes the agents order the wrong things. Other times, users want to complete the purchase themselves, or at least add an extra layer of security, rather than handing their credit card details directly to AI ([2]).
One concrete example from Business Insider: shopper Austin Welch asked Instinct for chicken wings and thighs, and eventually got 20 pounds (about nine kilograms) of chicken wings delivered to his door — an order Instinct had executed. Welch suspects the agent saw earlier larger grocery purchases in his history and concluded he wanted a similar quantity ([2]). The error was not in the payment, but in the agent inferring a detail — the quantity — that the customer never specified.
Amex's own trend data, from the Trendex report, shows the same divide. 66% of surveyed American consumers said they would likely use AI agents to find the best price in the next twelve months, and 61% for product or service recommendations, the company told TechCrunch ([1]). But when it comes to the actual purchase, willingness collapses: 52% would let an agent buy groceries or other low-priced items, while only 15% would allow more expensive purchases like electronics and appliances, according to Payments Dive's rendition of the Trendex survey ([3]). Earlier this year, Amex also said it would protect consumers from unintended AI purchases, subject to its own terms ([3]).
How the framework works
The details of the mechanism come from TechCrunch's interview with Bozian; Amex's own documentation is not available in the accessible source material. The flow has three steps ([1]):
-
Create Intent API. The framework starts with an API that captures the customer's purchase instructions and any constraints in the request — such as the merchant's name, maximum budget, and other product preferences. In practice, what the agent is asked to do is supposed to be formalized into machine-readable form before anything is purchased.
-
Intent ID and Proof of Intent Token. Next, an Intent ID is created — a unique identifier representing the intent — along with a Proof of Intent Token containing the customer's overall constraints.
-
Secure digital payment credential. The token is used to obtain a secure digital payment credential that the agent uses at checkout instead of the customer's actual credit card number.
That means the agent never sees the card number, and the payment credential it uses is bound to what the customer actually asked for — not to an open card the agent can use freely.
Why this differs from fixing things afterward
The payments industry's traditional response to wrong purchases is hindsight: dispute, refund, and chargeback when the customer discovers the error. Bozian claimed Amex's approach is different: "Our approach differs from others' implementations in that it pulls intent earlier into the transaction flow. And this is important because it helps mitigate some of the risk before the risk occurs," she told TechCrunch ([1]).
She gave an example: the customer asks the agent to buy white Nike sneakers in size 9, but the agent sends a request to buy shoes from Adidas. With the intent documented in advance, as part of the payment credential, the deviation between what was requested and what is being bought can be detected and stopped before execution — instead of being caught via a complaint after the packages have arrived.
Worth noting: Amex says that so far it has not seen an increase in disputes tied to agentic commerce, but that the company wanted to be proactive as this form of commerce scales ([1]). That is the company's own statement, not something that can be independently verified from the available sources — and it makes the framework a competitive strategic bet more than a response to documented harm.
The ambition and the ecosystem
Amex hopes the Intent Intelligence Framework will become an industry standard for AI-driven transactions, and says the company is already in discussions with partners across the payments ecosystem about implementing ACE, which includes agentic tokens with Amex's payment credentials and other parts of the developer toolkit ([1]). These are conversations — not signed agreements, named partners, or established adoption. Whether the framework actually becomes a standard depends on other card networks, agents, and merchants choosing to build on the same model.
What remains open
An honest reading of the announcement must include the following caveats:
- All technical detail about the framework comes via TechCrunch's interview and company statements; Amex's own documentation is not available in the source material ([1]).
- Adoption status beyond "discussions with partners" is unspecified ([1]).
- What was actually said at Bozian's Disrupt session is not reported in the sources ([1]).
- The connection between the framework and Amex's earlier pledge to protect consumers from unintended AI purchases is not elaborated ([3]).
- Amex's claim that there has been no increase in disputes is a company statement that cannot be independently verified ([1]).
What is interesting about the Intent Intelligence Framework is not just the token mechanism, but where it places responsibility: instead of letting agent developers invent their own security around card data, the card network makes the purchase instruction itself the authenticatable object. Whether that holds up in practice — and whether agent developers consider it worth implementing — remains unanswered for now.
Sources: [1] TechCrunch: Amex introduces a framework to prevent AI agents from making unauthorized purchases [2] Business Insider: AI Agents Instinct, Muse Redefine Personal Shopping Online [3] Payments Dive (via Yahoo Finance): Amex advises businesses on AI