Analysis
American Express wants a receipt trail before an AI agent, not a human, swipes the card. The company unveiled its "Intent Intelligence Framework" Tuesday at TechCrunch Disrupt 2026, detailed by TechCrunch, which captures and verifies a customer's purchase instructions -- merchant, maximum budget, product preferences -- before an AI shopping agent is allowed to complete a transaction.
How The Framework Actually Works
A Create Intent API records those constraints, the customer authenticates and approves them, and Amex issues an Intent ID along with a Proof of Intent Token holding the high-level rules the agent must follow. The agent then uses that token to pull a digital payment credential at checkout -- meaning the agent itself never handles the customer's actual card number. Amex's own example: if a customer asks an agent for white Nike sneakers in size 9 and the agent tries to buy Adidas instead, the mismatch between the token's constraints and the attempted purchase can block the transaction before it completes. The same Intent ID and token double as a fraud and dispute-resolution log, recording which agent authorized what purchase and when.
“## The Case For Building This Now Amex's own Trendex survey found 66% of U.S.”
Hannah Bozian, Amex's VP of Agentic Partnerships and Strategy, said the company hasn't yet seen a rise in disputes tied to agentic commerce -- this is a guardrail built ahead of the problem, not in response to one already showing up in Amex's dispute data. The framework builds on ACE, Amex's Agentic Commerce Experiences toolkit launched in April, extending a product line the company has been building out in stages rather than shipping as one single launch.
The Case For Building This Now
Amex's own Trendex survey found 66% of U.S. consumers say they're likely to use AI agents to find the best price over the next 12 months, and 61% said they'd use agents for product recommendations -- numbers that only matter if the unauthorized-purchase risk is real enough to need solving before, not after, agent-driven shopping scales. Because large language models are nondeterministic, the same prompt can produce a different purchase decision twice, which is the specific failure mode Amex is targeting rather than outright fraud.
What's missing from Amex's announcement: no competitor is named, and Visa, Mastercard, and Stripe -- all of which are building their own agentic-commerce infrastructure -- are absent from the framework entirely, meaning Amex's guardrail currently only protects transactions that run on Amex's own rails. Whether merchants and other card networks adopt anything resembling Amex's Intent ID model, or build incompatible versions of their own, is the open question that determines whether this becomes an industry standard or a single-issuer feature.