American Express is developing a new safeguard for AI-powered shopping that could help prevent artificial intelligence agents from making purchases their users never intended to authorize. The company’s new Intent Intelligence Framework is designed to verify a customer’s instructions before an AI agent completes a payment, adding another layer of protection to the growing agentic commerce ecosystem.
As AI assistants become more capable of researching products, comparing prices, and completing transactions, ensuring that they follow a user’s exact instructions is becoming increasingly important. Even when an agent misunderstands a request or selects the wrong product, the consequences can extend beyond an inconvenient shopping experience to an unwanted charge.
Amex aims to address this challenge by capturing a customer’s purchase intent early in the transaction process and carrying that information through checkout and other stages of the payment lifecycle.
What Is the Amex Intent Intelligence Framework?
The Intent Intelligence Framework is a payment security approach developed by American Express to help ensure that purchases made by AI agents match what customers have actually authorized.
Traditional online shopping generally involves a customer selecting a product and confirming the purchase. Agentic commerce changes this process by allowing an AI assistant to perform several steps independently, potentially including product discovery, selection, and payment.
However, AI models can misinterpret instructions or produce unexpected results. An assistant asked to purchase a particular product might select a different brand, exceed a budget, or overlook another important requirement.
Amex’s framework is designed to reduce these risks by capturing the customer’s original instructions before the transaction proceeds. The recorded intent can then be used to check whether the proposed purchase falls within the customer’s approved conditions.
According to American Express, the goal is to maintain the seamless, trusted, and secure payment experience customers expect while enabling AI agents to take on a greater role in shopping.
The company also hopes the framework will help establish a consistent approach to authorization across the agentic commerce industry, where consumers, merchants, payment providers, and AI agents must work together.
How Amex’s AI Purchase Protection System Works
The framework uses a series of steps to capture, authenticate, and reference a customer’s purchase instructions throughout the transaction.
1. The AI Agent Captures Purchase Intent
The process begins when a customer asks an AI agent to purchase something. The agent’s instructions can include details such as the product type, preferred brand, merchant, budget, size, and other shopping requirements.
Amex’s Create Intent API is designed to capture these instructions and the relevant restrictions attached to the request. The exact method of collecting intent may vary depending on the AI model or agent involved.
The framework is intended to remain flexible enough to support different agentic commerce implementations rather than relying on a single AI platform.
2. The Customer Authenticates and Approves the Instructions
Once the purchase intent has been captured, the customer authenticates their identity and approves the recorded instructions.
This step connects the transaction to the customer’s authorization rather than relying solely on the AI agent’s interpretation of the original request.
It also creates a reference point against which the proposed purchase can be checked before the payment moves forward.
3. Amex Generates an Intent ID and Proof of Intent Token
After the customer approves the captured intent, the system generates two important elements: an Intent ID and a Proof of Intent Token.
The Intent ID acts as a unique identifier for the customer’s approved purchase intent. It helps systems reference the authorization associated with a transaction.
The Proof of Intent Token contains high-level information about the customer’s purchasing restrictions. This information can be used to help ensure that the payment remains consistent with the approved request.
The token also supports the process of obtaining a secure digital payment credential for the AI agent to use at checkout instead of exposing the customer’s actual credit card number.
Together, these elements are designed to connect the customer’s instructions with the payment authorization process.
4. The Proposed Transaction Is Checked Against the Original Request
Before the purchase is completed, the framework can help determine whether the transaction matches the customer’s approved intent.
For example, imagine a customer instructs an AI assistant to buy white Nike sneakers in size 9. If the agent instead attempts to purchase Adidas shoes, the proposed transaction would not match the original request.
Amex says its framework could help prevent such a transaction from proceeding because the selected product conflicts with the customer’s authorized instructions.
The same principle can apply to other restrictions, such as a maximum spending limit or a specified merchant, depending on the conditions captured in the intent.
This approach is intended to identify certain errors before a payment is completed, rather than relying exclusively on fraud detection or a dispute after the purchase.
Why AI Shopping Needs an Additional Layer of Payment Security
AI-powered shopping is evolving from a tool that recommends products into a system that can potentially complete purchases with limited human involvement.
That shift creates new questions about payment authorization. An AI agent may understand most of a request correctly but still make a critical mistake when selecting a product or completing checkout.
For example, a customer might authorize a purchase only from a particular retailer, set a strict budget, or specify a particular product size. If an agent overlooks one of those requirements, the resulting transaction could fall outside the customer’s expectations.
Traditional payment security measures remain important, but verifying a customer’s original purchase instructions addresses a different part of the problem. A payment could involve valid credentials and still represent a purchase the customer did not intend to make.
Amex’s Intent Intelligence Framework is designed to bring that intent verification earlier into the transaction flow.
The company says this can help mitigate certain risks before they occur. The framework’s identifiers and authorization records may also help businesses investigate disputes by providing information about what was approved, which agent was involved, and when the authorization took place.
However, the framework should not be interpreted as a guarantee that every unauthorized transaction, AI error, or fraudulent payment will be prevented.
How Autonomous AI Agents Could Change Online Shopping
American Express believes purchase-intent verification could become increasingly valuable as AI agents gain more autonomy.
Today, many AI shopping systems request confirmation before completing a transaction. In the future, however, customers may allow agents to make certain purchases automatically within predefined limits.
Household essentials provide a straightforward example. A consumer might authorize an AI assistant to reorder paper towels or toilet paper whenever supplies run low, provided the purchase stays within a specified budget and meets basic product requirements.
In such a scenario, asking for manual approval every time could undermine the convenience of automated shopping. An intent-based verification system could instead help check whether each purchase satisfies the conditions the customer established.
The same principle could eventually support other recurring purchases and routine shopping tasks, although the level of automation and the safeguards used would depend on how individual services implement the technology.
The challenge for payment companies is to make these transactions convenient without losing the connection between a customer’s instructions and the purchases made on their behalf.
Amex Expands Its Agentic Commerce Experiences Developer Toolkit
The Intent Intelligence Framework builds on American Express’s Agentic Commerce Experiences initiative, also known as the ACE Developer Toolkit.
Introduced in April, the toolkit provides services intended to support AI-driven commerce and payment experiences. The new intent framework adds a mechanism for capturing and referencing the customer’s purchasing instructions as part of that broader initiative.
Amex is also developing agentic payment capabilities involving digital payment tokens that allow AI agents to use payment credentials without directly handling a customer’s actual card number.
The company says it is already discussing ACE implementation with partners across the payments ecosystem. These discussions cover the toolkit, its agentic payment tokens, and related capabilities.
American Express plans to continue developing ACE as the agentic commerce ecosystem evolves.
Wider adoption will depend on how payment providers, merchants, AI developers, and other participants integrate the technology into their existing systems.
Amex Says It Has Not Seen an Increase in Agentic Commerce Disputes
Despite the potential risks associated with AI-driven transactions, American Express says it has not yet observed an increase in disputes attributable to agentic commerce.
The company is developing the framework proactively as AI-assisted purchasing becomes more common, rather than responding to a reported surge in these disputes.
Consumer interest could help explain why payment providers are preparing for this shift. According to an Amex Trendex report cited by TechCrunch, 66% of surveyed U.S. consumers said they were likely to use AI agents to research the best price for a product or service over the following 12 months.
The same report found that 61% said they were likely to use AI agents for product or service recommendations during that period.
These figures indicate interest in using AI for shopping-related tasks, but they do not establish how many consumers are already comfortable allowing agents to complete purchases independently.
For American Express, the findings highlight the importance of building trust and clear authorization mechanisms before autonomous shopping becomes more widespread.
What the Intent Intelligence Framework Means for Consumers and Merchants
If adopted broadly, intent-based payment controls could offer benefits to several participants in the commerce ecosystem.
For consumers, the framework is designed to help ensure that AI agents follow their purchasing instructions, including relevant product preferences and spending restrictions.
For merchants, a consistent method of referencing customer authorization could help support transaction handling and the investigation of certain payment disputes.
For payment providers, intent information could provide additional context for fraud prevention, payment authorization, and post-transaction reviews.
For AI developers, the framework could offer a way to incorporate purchase-intent verification into agentic commerce systems without depending on one particular AI model.
Nevertheless, the practical impact will depend on implementation. The accuracy of the captured intent, the quality of transaction checks, and the ability of participating systems to exchange the relevant information will all matter.
The framework also does not eliminate the need for established payment security controls, customer authentication, or appropriate dispute-resolution procedures.
What’s Next for American Express’s AI Payment Security Plans?
American Express intends to continue developing its Agentic Commerce Experiences platform as AI-driven shopping expands.
On October 13, 2026, Hannah Bozian, Amex’s vice president of Agentic Partnerships and Strategy, was scheduled to discuss the Intent Intelligence Framework at TechCrunch Disrupt 2026 on the Industry Stage. The discussion was expected to provide further insight into the framework and its implications for agentic commerce.
The next phase will involve how the technology is adopted and implemented across the payments ecosystem. American Express has indicated that it is engaging with industry partners, but broad availability and adoption should not be assumed without further confirmation.
As AI agents take on more responsibility for purchasing decisions, the ability to verify what a customer actually authorized could become an important part of digital payment security.
Amex’s Intent Intelligence Framework represents an effort to address that challenge by connecting a customer’s instructions to the transaction before the payment is completed.
Whether it becomes a widely adopted industry standard will depend on implementation, interoperability, and the willingness of payment providers and AI developers to incorporate intent verification into their systems.
