American Express Launches New Framework to Secure AI Agent Transactions

As personal AI agents increasingly manage transactions on behalf of users, the non-deterministic nature of large language models raises concerns about potential errors or unauthorized actions. This creates a risk where AI bots might misfire and execute purchases that were not explicitly intended.

To address this challenge, American Express is launching a new framework aimed at capturing and authenticating a customer's purchase intent prior to an AI agent finalizing a transaction. The company emphasizes the importance of ensuring purchase instructions are followed throughout the entire transaction lifecycle. Dubbed the "Intent Intelligence Framework," Amex hopes this solution will set an industry standard for secure AI-powered commerce.

"These new transactions will involve a consumer, a merchant, and an agent," explained Hannah Bozian, Amex's Vice President of Agentic Partnerships and Strategy, in an interview with TechCrunch. She noted that the company is committed to maintaining seamless, trusted, and secure experiences for its card members.

This release comes shortly after Amex introduced the Agentic Commerce Experiences (ACE) Developer Toolkit in April, a suite of services that now incorporates the intent framework. Although Amex has not yet observed a spike in disputes related to agentic commerce, the company is taking a proactive approach to secure the technology as this sector expands, driven in part by growing consumer interest.

According to an Amex Trendex report cited by TechCrunch, 66% of surveyed U.S. consumers are likely to use AI agents to research product prices, and 61% intend to use them for recommendations within the next year.

The new intent framework begins with a Create Intent API designed to capture the customer's purchase instructions and constraints, such as merchant name, budget limits, and product preferences. While the method of collection may vary based on the AI model used, Bozian stated that the framework is being designed to remain flexible enough to support various agentic commerce models.

Following authentication, the customer reviews and signs off on the captured intent. An Intent ID is then generated, along with a Proof of Intent Token containing the customer's high-level constraints. This token is used to generate a secure digital payment credential for the agent to use at checkout, ensuring the transaction remains secure and aligned with the customer's authorization while keeping the actual credit card number private.

Amex highlighted that these Intent IDs and tokens can be utilized throughout the transaction process to assist with fraud prevention and dispute resolution.

"Our approach differs from others in that it pulls intent earlier in the transaction flow, which is crucial for mitigating risk before it occurs," Bozian explained. She offered the example of a customer requesting white Nike sneakers in size 9; if an agent attempts to purchase shoes from Adidas instead, Amex would block the transaction to prevent the error.

Bozian also noted that while current agents often require consent before purchasing, Amex anticipates a future with more autonomous agents. For example, consumers might allow agents to autonomously reorder household staples where the primary constraints are price and size rather than specific brands. In these cases, having technology in the loop to verify intent is essential.

Maintaining a record of authorized transactions by specific agents could help address agent-related fraud, she added. Amex is currently in discussions with partners across the payments ecosystem to implement ACE, including the tokens containing Amex payment credentials. The company plans to continue developing ACE as the agentic commerce landscape evolves.