Personal artificial intelligence agents are increasingly stepping into the role of digital personal assistants, browsing the web, comparing prices, and executing transactions on their users’ behalf. However, given the inherently nondeterministic nature of large language models, AI bots are prone to unpredictable behavior and hallucinations—doing things they were never actually instructed to do. This technical reality raises significant financial risks, chief among them being the possibility of AI misfiring and executing unauthorized or erroneous transactions that could leave consumers financially exposed.
To get ahead of this burgeoning industry problem, American Express is introducing a robust new framework designed to capture and authenticate a customer’s exact purchase intent before an AI agent is ever allowed to complete a transaction. The core philosophy behind the initiative is to establish a reliable mechanism ensuring that a consumer’s specific purchase instructions are accurately carried throughout the entire transaction lifecycle, from the initial product search all the way through to final fulfillment and post-purchase management.
Dubbed the "Intent Intelligence Framework," the financial services giant hopes the new system will eventually become a trusted industry standard for AI-powered transactions. The release of this framework follows closely on the heels of the company’s April launch of Amex’s Agentic Commerce Experiences, also known as the ACE Developer Toolkit. The toolkit packages a handful of specialized services for developers, which now officially incorporate the new intent framework.
Industry Context and Consumer Readiness
American Express notes that while it has not yet witnessed a tangible spike in customer disputes directly attributable to agentic commerce, the company wanted to be proactive in building out foundational security infrastructure as this type of commerce rapidly scales. This forward-looking stance is largely driven by clear signals from consumers, who have demonstrated a strong willingness to embrace AI-driven shopping tools.
According to a recent Amex Trendex report, an overwhelming 66 percent of surveyed U.S. consumers indicated they are likely to use AI agents to research the best prices for products or services over the coming year. Furthermore, 61 percent reported that they are likely to lean on AI agents for personalized product or service recommendations during the same timeframe. As consumer comfort with automated shopping assistants grows, the financial ecosystem must evolve to match that shift with adequate safeguards.
"These new transactions will have a consumer, a merchant, and an agent," explained Hannah Bozian, Amex’s VP of Agentic Partnerships and Strategy, in an interview. She emphasized that the company’s primary objective is to ensure that agentic transactions remain "seamless, trusted, and secure, the way our card members expect them to be."
How the Intent Framework Operates
At a technical level, Amex’s new intent framework begins with a specialized Create Intent API. This application programming interface is designed to capture the customer’s explicit purchase instructions along with any auxiliary constraints they may have included in their request, such as a specific merchant’s name, a strict maximum budget, and other precise product preferences. While the exact method of how this intent is collected may vary depending on the underlying AI model involved in the transaction, the overarching plan is to make the framework flexible enough to support all major agentic commerce models currently in development.
Once the parameters are set, the customer authenticates their identity and formally signs off on the intent captured by the agent. Following this authorization, a unique Intent ID is generated to represent the specific intent, alongside a Proof of Intent Token that encapsulates the customer’s high-level constraints and parameters.
Crucially, this token is subsequently used to obtain a secure digital payment credential. The AI agent uses this temporary, secure credential at checkout instead of exposing the customer’s actual credit card number to the merchant or the digital ecosystem. In doing so, Amex explains, the purchase remains highly secure while strictly aligning with what the customer originally authorized the agent to do. Beyond the checkout phase, the Intent ID and token can be utilized across different parts of the broader transaction process, assisting companies in managing critical areas such as fraud prevention and dispute resolution.
Mitigating Risk Before It Occurs
By shifting the validation phase earlier in the transaction pipeline, Amex aims to intercept errors before they manifest as financial losses or customer service headaches. Traditional payment security often focuses on intercepting fraud after a transaction has already been submitted, but agentic commerce introduces the novel risk of the buyer’s own agent making an operational mistake.
"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. To illustrate the concept, she offered a relatable hypothetical scenario: a customer instructs their personal AI agent to purchase a pair of white Nike sneakers in a specific size 9, but the agent mistakenly submits a purchase request for shoes from Adidas instead. In such a case, Amex’s framework would have the capability to intercept the error and prevent the transaction from going through because the agent deviated from the defined parameters.
Bozian also pointed out that while many consumer-facing agents today are programmed to repeatedly ask for manual customer consent before making any final purchase, the company foresees a future where agents will act with much higher degrees of autonomy. For instance, a consumer might grant their AI agent standing permission to automatically re-order household staples, such as paper towels or toilet paper—product categories where the primary constraints are limited strictly to price thresholds and package sizes rather than specific brand loyalty or merchant preference. In these automated scenarios, having specialized technology continuously in the loop to perform real-time validation checks becomes essential to ensuring that the completed transactions accurately match the customer’s underlying intent.
Furthermore, maintaining an immutable log of what was authorized by which agent, and at what exact time, could provide significant utility in addressing disputes or suspected fraud originating on the agent’s side of the transaction, Bozian noted.
Ecosystem Collaboration and Future Outlook
American Express is already actively engaged in discussions with various partners across the broader payments ecosystem regarding the implementation of the ACE toolkit. This includes integrating agentic tokens containing Amex payment credentials alongside other foundational components of the developer kit. As the agentic commerce ecosystem continues to mature and expand, the company intends to iteratively develop ACE further to meet the evolving demands of developers, merchants, and cardholders alike.
Industry stakeholders will get a closer look at these developments later this year. On October 13 at TechCrunch Disrupt 2026’s Industry Stage, Hannah Bozian, VP of Agentic Partnerships and Strategy at American Express, is scheduled to discuss the company’s new "Intent Intelligence Framework" in greater detail, exploring what it ultimately means for the rapidly unfolding world of agentic commerce.
