On 13 October 2026 American Express published its Intent Intelligence framework — its recommended approach to capturing and authenticating a customer’s purchase intent before an AI agent completes a transaction, and then preserving that intent as a control through the transaction lifecycle.
What happened
Amex frames the shift plainly: in traditional payments a person browses, decides and executes each step; with agents, a user may state what they want once, and the agent “interprets those instructions, evaluates changing conditions, and ultimately executes a purchase.” That raises a KYA question the card networks care about directly — did the customer actually authorize this agent action, and within what limits? — and a commercial one: disputes, which issuers are responsible for managing.
How it works. A user expresses constraints and delegates final purchase authority to an agent. Amex defines two intent shapes and two cart shapes:
- Broad intent — e.g. “build me a website, but don’t spend more than $100.”
- Specified intent — e.g. “if the sneakers I want are restocked before April 15, buy one pair in a US women’s size 8 for $200 or less.”
- Order-ready cart — the agent fills a cart and the user signs off before the agent completes the purchase.
- Proposed cart — the user confirms a cart the agent presents; the agent then fills and checks out.
The user must authenticate their identity and sign off on the intent the agent captured. Sign-off can be via passkey, or the issuer may delegate sign-off to a third party such as the agent or a wallet provider (issuers “may require incremental data to support sign-off delegation“). After authentication, Amex issues two artifacts:
- an Intent ID — the unique key for the intent record and any later lifecycle changes; and
- a Proof of Intent Token — a JSON Web Token representing the high-level constraints of the intent.
A Create Intent API shares the confirmed intent (merchant, maximum budget, preferences, and more) to create the Proof of Intent Token, which can then be used to obtain a scoped payment credential and complete the transaction. Per the user’s constraints, the token expires on the user-set date and can be single-use or multi-use. Amex also defines data elements the agent shares with the issuer, including max amount, intent expiry, merchant data, item-level conditions, maximum number of transactions, order context and purchase criteria.
Why issuers. Because issuers manage cardholder disputes, Amex argues that giving them authenticated purchase-intent information before the agent transacts lets them assess dispute risk up front and later evaluate a dispute against the intent the cardholder confirmed. The framework also defines ecosystem roles — agent, vault provider, digital wallet, and issuer or network — and says intent can be updated or cancelled throughout its lifecycle.
Why it matters for KYA
This is a payments network writing down how it wants delegated authority to work, and it lands squarely on the KYA problem: an agent acting with someone else’s money needs a verifiable link between the agent’s action and the principal‘s authorization. Intent Intelligence is that link, in three moves that map cleanly onto KYA vocabulary:
- Delegation made explicit. “Broad” and “specified” intent turn a vague instruction into a bounded mandate (max amount, merchant, expiry, item conditions) rather than an open-ended permission.
- Scoped credentials. The Proof of Intent Token is a JWT carrying the constraints, used to obtain a scoped payment credential — a spend-limited, expiry-bounded, single- or multi-use credential rather than a reusable card number handed to an agent.
- Accountability and revocation. The Intent ID is a lifecycle record that can be updated or cancelled, and it is retained so an issuer can check a later dispute against the confirmed intent — the audit half of KYA, applied to agent payments.
Two honest caveats. First, this is one issuer’s framework, published for its ecosystem; it is not a cross-industry standard, and it sits alongside — not in place of — protocol efforts such as Google’s AP2 mandates (which the Chain of Trust demo above also uses). Interoperability across issuers, wallets and agents is the open question. Second, the security of the whole scheme rests on the sign-off step: if an issuer delegates sign-off to the agent, the KYA guarantee degrades toward the agent trusting itself — which is precisely the kind of delegation a KYA control must be able to bound and audit.
For a practitioner, the takeaway is concrete: authenticated, scoped, expiring purchase intent is becoming a first-class control in agent commerce, and “who authorized this, and within what limits?” is now something a card network expects to be able to answer before an agent pays.
Sources
- American Express, “American Express Publishes Intent Intelligence Framework for Trusted Agentic Commerce” (newsroom, 13 October 2026) — americanexpress.com.
- American Express Technology, “Intent Intelligence: Building Trust in Agentic Commerce” (technical framework describing Intent ID, the Proof of Intent Token, the Create Intent API and ecosystem roles) — americanexpress.io.
- TechCrunch, “Amex introduces a framework to prevent AI agents from making unauthorized purchases” (13 October 2026) — techcrunch.com.
- Amex’s related Agentic Commerce Experiences (ACE) Developer Kit and merchant playbook material — americanexpress.com.
Note: details and terminology (Intent ID, Proof of Intent Token, Create Intent API, ecosystem roles) are American Express’s own; this post describes the published framework, not a measured security property of any live deployment.
Know Your Agent (KYA) explains agent identity, verification and accountability. This is an explainer, not legal or compliance advice — see our Sources & methodology. New to KYA? Start with What is Know Your Agent? and the glossary.