American Express Launches Intent Intelligence Framework to Prevent Unauthorized AI Agent Transactions

According to foreign media reports, American Express has officially launched its "Intent Intelligence Framework," embedding a consumer purchase intent verification step into the full transaction process executed by AI agents, mitigating the risk of unauthorized transactions arising from the non-deterministic characteristics of large language models. This framework is a new addition to American Express's ACE (Agent Commerce Experience) developer toolkit, first introduced in April this year. The company aims for this mechanism to become an industry-wide standard for AI-driven transactions, ensuring agent-based transactions meet cardholders' expectations for a seamless, trustworthy, and secure experience. The framework operates starting from the Create Intent API, which captures consumers' specific purchase instructions along with constraints such as merchant names, maximum budget, and product preferences set by users. The data collection method adapts to the operational logic of different AI models, maintaining sufficient flexibility to cover various agent commerce models. After consumers complete identity authentication and confirm the purchase intent captured by the AI, the system generates a unique intent ID representing that intent, along with an intent attestation token containing the user's core constraints. During the checkout phase, the AI agent uses this token to obtain a secure digital payment credential, without needing to access the consumer's actual credit card number, ensuring payment security while guaranteeing that transaction content falls within the user's authorized scope. The generated intent ID and token can be used throughout the entire transaction process, supporting fraud prevention, dispute resolution, and other aspects. The key difference between this framework and existing industry solutions lies in front-loading the intent verification step to the early stages of the transaction flow, completing risk mitigation before issues arise. If a consumer requests an AI to purchase a pair of white Nike sneakers in size 9, and the AI erroneously submits a purchase request for Adidas shoes, the system can directly block the transaction. While most current AI agents seek consumer consent before payment, future AI agents are expected to operate with higher autonomy, allowing consumers to authorize agents to automatically reorder household items like paper towels and toilet paper—scenarios where users typically set only price and specification constraints without limiting merchant or brand. In such cases, the additional verification ensures transactions adhere to user-defined rules, and the full-chain authorization logs provide traceability for potential fraud on the agent side. American Express has not yet observed a rise in dispute volumes related to agent commerce but is proactively introducing this technology to align with the expansion pace of such transactions. Its recent Trendex report shows that 66% of surveyed U.S. consumers are likely to use AI agents to research optimal prices for products or services in the next 12 months, and 61% may rely on AI for product or service recommendations. Currently, American Express is in discussions with multiple partners across the payment ecosystem to roll out the ACE toolkit, covering agent tokens embedded with American Express payment credentials and other components of the developer suite. The company will continue to iterate on related features as the agent commerce ecosystem evolves. At an industry forum held on October 13, Hannah Bozian, Vice President of Agent Partnerships and Strategy at American Express, shared detailed insights into the new framework and the company's roadmap for agent commerce.

This article was first published on the official website of Ebrun.