The framework is designed to authenticate what customers authorize AI agents to buy, carry those constraints into payment execution and help issuers manage fraud and disputes.
American Express has published its Intent Intelligence framework, outlining how payment providers could verify a customer’s purchase instructions before an artificial intelligence (AI) agent completes a transaction. The approach is designed to preserve those instructions throughout the payment process, helping ensure that an agent’s actions remain within the boundaries approved by the customer.
Announced on 13 October 2026, the framework is available through the American Express Agentic Commerce Experiences (ACE) Developer Kit. It introduces a structured way to capture purchase requirements, authenticate the customer’s approval and connect that approval to payment credentials used by an AI agent. The objective is to give issuers earlier visibility into what a customer has authorized, rather than relying solely on information presented when a payment request arrives.
The announcement addresses a central challenge for agentic commerce: when software can shop and pay on a person’s behalf, how can merchants, payment providers and customers establish what was actually authorized?
American Express says its approach is intended to help manage that challenge by making authenticated purchase intent available earlier in the transaction lifecycle. The company also sees the framework as a contribution towards a more consistent industry approach to trusted AI-driven commerce.
Why Purchase Intent Matters in Agentic Commerce
Traditional e-commerce generally involves a customer browsing products, reviewing a basket and completing a checkout. Even when automated recommendations influence the decision, the customer typically remains involved in the final purchase steps.
Agentic commerce changes that sequence. A customer might ask an AI agent to find a flight below a specified price, purchase an item when it becomes available again or order supplies within a defined budget. The agent may then compare options, select a merchant and complete the transaction when the customer’s conditions are met.
This introduces a new control problem. The agent must interpret the customer’s instructions correctly, respect any limits and execute the purchase without changing important details along the way.
A customer might authorize a purchase up to a maximum amount, for example, but the final transaction could differ from the original request if the agent selects the wrong product, changes the basket or misinterprets a condition. In other cases, a customer could dispute a transaction after the event because the purchase does not match their understanding of what they approved.
Intent Intelligence is designed to address this gap by representing the customer’s instructions as structured information that can be authenticated and used in later payment steps.
The distinction is important: the framework is not simply about confirming that a person logged in or approved access to a payment card. It aims to connect that authenticated approval to the parameters of a particular purchase or purchasing task.
How American Express Intent Intelligence Works
American Express describes a process built around capturing intent, authenticating approval, creating proof of that intent and using it to constrain payment credentials and subsequent transaction activity.
1. Capture the customer’s instructions
The process begins with the Create Intent API. It captures the customer’s purchase instructions and constraints, which can include the intended merchant, maximum budget and other preferences.
An AI agent may need to ask follow-up questions before the instructions are sufficiently clear. For example, a customer asking an agent to buy a product when it becomes available may need to specify the required model, size, price ceiling and purchase deadline.
The customer then authenticates their identity and signs off on the intent captured by the agent. This step is intended to establish that the structured instructions reflect what the customer has authorized.
The precise way intent is collected and confirmed may differ depending on the agent, wallet or other parties involved in a particular implementation.
2. Create an Intent ID and Proof of Intent Token
After the intent is authenticated, the framework creates two important elements.
The Intent ID is a unique identifier for the customer’s intent record. It can be used to reference that record in subsequent requests and lifecycle operations.
The Proof of Intent Token represents the high-level constraints associated with the approved intent. American Express describes it as a JSON Web Token, or JWT, that carries information about the boundaries of the authorized purchase.
These elements provide a way for systems involved in a transaction to refer back to the customer’s approved instructions. Rather than treating the purchase as an isolated payment request, the framework is designed to link it to an authenticated intent record.
3. Use a scoped digital payment credential
The Proof of Intent Token can be used to obtain a secure digital payment credential for the agent to use at checkout, instead of exposing the customer’s actual card number.
The credential can be scoped according to the customer’s approved intent. Depending on the use case and implementation, the constraints may include a maximum spend, merchant preferences or an expiry date.
The framework’s technical description also explains that an intent can be configured for single use or for multiple uses, including transactions involving one or more merchants where the approved conditions permit it.
This is a significant part of the proposed control model. The goal is not merely to authenticate an agent or replace a card number with a digital credential. It is to connect the payment credential to the customer’s approved purchasing boundaries.
The framework does not mean every AI agent automatically receives unrestricted access to a customer’s card. Rather, it sets out an approach for linking authorized intent to payment credentials, subject to the relevant implementation and issuer controls.
4. Carry intent through the transaction lifecycle
The Intent ID and Proof of Intent Token can serve as controls in later stages of the purchase journey. They are intended to help validate whether a payment request aligns with the customer’s authorization.
American Express says the framework can support fraud prevention, authorization decision-making and dispute resolution. For example, an issuer could use information about the approved purchase when assessing a payment request or reviewing a subsequent dispute.
The company’s technical explanation also describes the ability to create, update or cancel intent. This matters because some agentic purchases depend on future events or changing conditions. Customers may need to modify their instructions or withdraw permission before an agent completes a purchase.
In practice, the extent of these controls will depend on how agents, wallets, issuers and other participants integrate the framework and share the relevant information.
What the Framework Could Mean for Fraud and Disputes
The growth of agentic commerce raises questions about responsibility when an AI agent makes a mistake. A payment may be technically authenticated, yet the customer could still argue that the product, price or merchant was not what they authorized.
Intent Intelligence aims to give payment providers additional context before a transaction occurs. If an issuer can compare a payment request with authenticated purchase constraints, it may be better positioned to identify inconsistencies before authorizing a transaction.
That information could also be useful after the event. A record of the approved intent may help the issuer investigate whether a disputed purchase matched the customer’s instructions.
This could become increasingly relevant as AI agents take on more complex tasks, including purchases triggered by price changes, product availability or recurring business needs. Such tasks can involve delegated authority that remains active over a period of time, rather than a single click at checkout.
However, the framework should not be interpreted as a guarantee that fraud or disputes will disappear. An authenticated intent record is one potential source of evidence and one layer of control. It does not, by itself, establish that an AI agent interpreted instructions correctly, that every merchant detail is accurate or that a payment will always be blocked when it falls outside a customer’s expectations.
Its effectiveness will depend on implementation quality, the accuracy of the information captured, how the constraints are enforced and how participating systems handle exceptions.
The Role of the ACE Developer Kit
Intent Intelligence is a core service of the American Express Agentic Commerce Experiences Developer Kit, introduced in April 2026. The broader ACE framework is intended to help developers integrate American Express payment capabilities into AI-powered experiences.
The developer kit includes a wider set of capabilities related to agent registration, account enablement, intent, payment credentials and cart context. Together, these are intended to help participants build transactions with greater visibility and control as AI agents take on more purchasing tasks.
American Express occupies a distinctive position in this market because it operates as both a card issuer and a payment network. That gives the company visibility into aspects of the transaction that may otherwise be split across separate organisations.
The company argues that this position can help it connect customer authorization with payment processing, risk assessment and dispute management. It is also developing ACE integrations as the wider agentic commerce ecosystem evolves.
The availability of a published framework gives developers and other participants a technical reference for considering how purchase intent could be represented and carried between systems. It does not mean every ACE service or integration is universally available, nor that all agentic commerce platforms currently support the same controls.
American Express Seeks Interoperable Agentic Payment Standards
Intent Intelligence is also part of a broader effort to establish common approaches to secure agentic commerce.
American Express says it is engaging with EMVCo to support secure card-based agentic transactions and supporting the x402 Foundation’s work on standards for internet-native payments.
These initiatives address a wider industry challenge. AI agents may operate across different merchants, wallets, payment networks and software platforms. If every participant defines customer intent and agent authorization differently, the ecosystem could become difficult to integrate and harder to secure consistently.
EMVCo’s work is relevant to card-based payments, where participants need interoperable ways to manage delegated authority and assess transactions. The x402 Foundation focuses on internet-native payment mechanisms. Their approaches address different parts of a rapidly developing market, so they should not be treated as interchangeable standards.
The publication of American Express’s framework is a company-led contribution to this evolving landscape. It is not, by itself, proof that the industry has adopted one universal standard for agentic commerce.
Broader interoperability will depend on technical alignment, participation from payment providers and developers, and agreement about how customer authorization should be represented, transmitted and enforced.
What Merchants, Issuers and Developers Should Watch
For merchants, the growth of agentic commerce could change how purchases are initiated and completed. A merchant may interact with an AI agent rather than directly with a customer at every stage. Reliable purchase context could help clarify whether an order matches the customer’s approved conditions.
For issuers, authenticated intent could become another input in payment authorization and risk management. It may also provide useful evidence when a customer challenges an agent-driven transaction. The key question will be whether the information is available in time and in a form that can be reliably evaluated.
For developers and wallet providers, the framework highlights the importance of treating intent as a lifecycle object rather than a one-time instruction. Permissions may need to expire, change or be revoked, while payment credentials should remain consistent with the approved constraints.
For customers, the intended benefit is greater control over what an AI agent can buy and under which conditions. But meaningful control requires more than a confirmation screen. Customers need understandable permissions, clear limits and practical ways to review or cancel an authorization.
Across the ecosystem, implementation details will matter. Participants will need to consider privacy, secure storage, token handling, identity authentication, error resolution and the responsibilities of each party when an agent acts outside its authorization.
The Road Ahead for Trusted Agentic Commerce
American Express’s Intent Intelligence framework addresses one of the fundamental questions raised by autonomous shopping: how can a payment system distinguish between a customer’s general request and the specific purchase the customer actually approved?
By capturing intent before an agent executes a transaction and linking that intent to subsequent payment controls, the framework offers a model for carrying customer authorization beyond the initial interaction.
Its wider significance will depend on adoption and execution. Payment providers, merchants, AI developers and wallet operators will need to decide how to exchange intent data, enforce limits and handle situations where the agent’s actions diverge from the customer’s instructions.
The publication of the framework provides a reference point for that work, while American Express’s engagement with industry organisations signals that interoperability remains a priority.
The bottom line: trusted agentic commerce will require AI agents to do more than complete transactions efficiently. They must also operate within clear, authenticated boundaries. American Express is proposing one way to connect those boundaries to the payment process, but the practical impact will depend on how broadly and consistently the approach is implemented.


Comments
No comments yet. Be the first to share your thoughts.
Leave a comment