Amex's pitch is simple: right now, nothing forces an AI shopping agent to prove it understood you. The Intent Intelligence Framework, released as an extension of the Agentic Commerce Experiences (ACE) Developer Toolkit Amex launched in April, tries to close that gap by moving verification earlier in the transaction — before checkout, not after a dispute.
How It Actually Works
A Create Intent API captures the customer's instructions and constraints — merchant, budget ceiling, size, brand. The customer authenticates and approves what was captured. The system then generates an Intent ID and a Proof of Intent Token encoding those constraints, which the agent must use to obtain a digital payment credential — meaning the agent never touches the customer's actual card number. Amex's own example: a shopper asks for white Nike sneakers, size 9; if the agent tries to check out with Adidas instead, the mismatch gets caught before the transaction clears, not after.
Seamless, trusted, and secure.
— Hannah Bozian, Amex VP of Agentic Partnerships and Strategy, on the framework's goal
Bozian frames Amex's differentiation as pulling intent verification earlier than competitors do — reducing risk before it occurs rather than adjudicating it afterward. The Intent ID and token are designed to stick around after checkout too, supporting fraud prevention and dispute resolution across the life of the transaction.
Why Now
Amex says it hasn't actually seen a rise in agentic-commerce disputes yet — this is a pre-emptive move, not a response to a crisis. But the numbers explain the urgency: in Amex's own Trendex research, 66% of U.S. consumers say they're likely to let an agent hunt for the best price, and 61% say they'll lean on one for recommendations within a year. That's the same adoption curve this newsletter has been tracking all month, now meeting its first real accountability infrastructure. Amex is in active discussions with payments-ecosystem partners about implementing ACE more broadly, which is the real test of whether this becomes a shared standard or stays an Amex-only safeguard.
What This Means for CPG Brands
The Nike/Adidas example is really a brand-substitution story, and it should get CPG leaders' attention both ways. On the upside, a verified intent token is the first mechanism that could stop an agent from silently downgrading a named brand to a cheaper alternative at the moment of purchase — something brands have had zero visibility into, let alone control over, until now. On the downside, Amex also describes a future where agents handle routine reorders — household staples — more autonomously, checking only price and size. In that world, a brand that isn't the structured, legible default in an agent's intent record doesn't get to make its case at all; it just loses the reorder quietly, every time. Being verifiable to an intent API is shaping up to be as commercially important as being visible in search once was.
The Bottom Line
Trust frameworks for agentic commerce have mostly been promises so far — playbooks, liability pledges, voluntary standards. This is the first one with actual API mechanics attached, from a payments network with the leverage to make it stick. Whether it becomes the industry standard Amex wants depends entirely on whether merchants and other agent platforms adopt it — but for brands, the lesson arrives regardless: the gap between being visible to an agent and being verifiably what the shopper meant is about to become a real line item.
Sources: TechCrunch, "Amex introduces a framework to prevent AI agents from making unauthorized purchases," Oct. 13, 2026; Yahoo Finance / American Banker, "American Express Is Telling Merchants How to Prepare for Agent Shopping," Oct. 7, 2026.