Amex Builds Guardrails to Stop AI Agents From Buying the Wrong Shoes
Amex's new "Intent Intelligence Framework" authenticates purchase intent before an AI agent checks out, aiming to make agentic commerce secure — and to set the industry standard.
By Olivia Hart
4 min read
Updated
What's News
- Amex launched the "Intent Intelligence Framework" to authenticate purchase intent before an AI agent completes a transaction.
- 66% of surveyed U.S. consumers said they are likely to use AI agents for price research; 61% for product recommendations within 12 months.
- The framework uses a Create Intent API, an Intent ID and a Proof of Intent Token to replace the customer's real card number at checkout.
- The intent framework ships inside the ACE Developer Toolkit, which Amex launched in April.
- VP Hannah Bozian will detail the framework on October 13 at TechCrunch Disrupt 2026's Industry Stage.
By Olivia Hart, Business Bearings
American Express has launched a framework that captures and authenticates a customer's purchase intent before an AI agent completes a transaction, an attempt to stop erratic bots from spending cardholders' money. The company calls it the "Intent Intelligence Framework," and it wants the rest of the payments industry to adopt it as a standard.
The timing matters more than the naming. Amex said it has not yet seen a rise in disputes tied to agentic commerce, according to its own account — it is building the technology ahead of the problem, not in response to one. Whether the demand materializes at the scale Amex expects remains the open question.
Why is Amex doing this now?
Large language models are nondeterministic: they sometimes do what they were not asked to do. Applied to payments, that failure mode means unauthorized purchases made at machine speed.
Amex says consumers are signaling readiness anyway. In its recent Trendex report:
- 66% of surveyed U.S. consumers said they are likely to use AI agents to research the best price for a product or service.
- 61% said they are likely to use an agent for product or service recommendations within the next 12 months.
Consumer intent surveys, it should be noted, measure stated willingness — not completed transactions. The takeaway: Amex is betting on a behavior curve before the data confirms it.
How does the framework actually work?
The system starts with a Create Intent API. It records the customer's purchase instructions and any constraints bundled into the request — merchant name, maximum budget, product preferences. How that intent gets collected will vary by AI model; Bozian said the framework is designed to be flexible enough to support all agentic commerce models.
From there, the process runs in four steps:
- The customer authenticates their identity and signs off on the intent the agent captured.
- Amex generates an Intent ID, a unique identifier representing the intent.
- A Proof of Intent Token — carrying the customer's high-level constraints — is created alongside it.
- That token obtains a secure digital payment credential, which the agent uses at checkout instead of the customer's real card number.
The Intent ID and token also plug into fraud prevention and dispute resolution, Amex said. In plain words: every agentic purchase now carries a verifiable record of who authorized what, and when.
What does it change in practice?
Hannah Bozian, Amex's VP of Agentic Partnerships and Strategy, framed the stakes in an interview with TechCrunch: "These new transactions will have a consumer, a merchant, and an agent." She said Amex wants agentic transactions to remain "seamless, trusted, and secure, the way our card members expect them to be."
The differentiator, she argued, is timing. "Our approach differs from the implementations of others in that it pulls intent earlier into the transaction flow. And this is important because it helps mitigate some risk before the risk occurs," Bozian said.
Her example: a customer asks an agent to buy white Nike sneakers in a size 9; the agent submits a request for Adidas shoes instead. Under the framework, Amex could block the transaction because the agent erred — catching the mistake before money moves, not after a dispute is filed.
What about fully autonomous agents?
Bozian acknowledges that many agents today ask for consent before each purchase. Amex is designing for a future where they do not. A consumer might let an agent auto-reorder household staples — paper towels, toilet paper — where the only real constraints are price and size, not brand or merchant.
That is precisely where the risk grows, and where an automated check in the loop earns its keep. Bozian also noted that a log of what was authorized, by which agent and when, could help address fraud committed on behalf of the agent itself.
What comes next?
The framework ships as part of Amex's Agentic Commerce Experiences, the ACE Developer Toolkit launched in April, which now includes the intent framework among its services. Amex said it is already in discussions with partners across the payments ecosystem to implement ACE, including the agentic tokens holding Amex payment credentials.
The company said it plans to keep developing ACE as agentic commerce evolves — a deliberately open-ended commitment with no named partners or rollout dates attached. Bozian will present the framework in detail on October 13 at TechCrunch Disrupt 2026's Industry Stage; expect the hard questions about adoption, liability and interchange economics to surface there, because Amex has so far answered none of them.
The so-what: if Amex's framework becomes the de facto standard, it positions the network as the trust layer for AI-driven spending — a role worth far more than any single transaction it blocks.
Source: TechCrunch
More from Olivia Hart
Show full bio
Staff writer covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
727 articles