As personal artificial intelligence agents increasingly step into the role of digital personal assistants, they are beginning to make financial transactions directly on behalf of their users. However, this growing shift toward agentic commerce introduces a distinct set of technological challenges. Chief among them is the inherently nondeterministic nature of large language models, which can occasionally cause AI bots to hallucinate, misunderstand prompts, or execute actions they were never programmed or asked to do. In the realm of financial technology, this unpredictability raises significant concerns regarding AI misfires and unauthorized purchases.
To get ahead of this problem before it escalates into a widespread industry crisis, American Express is introducing a new framework designed to capture and authenticate a customer’s purchase intent long before an AI agent actually completes a transaction. The overarching idea is to ensure that a customer’s specific purchase instructions are faithfully carried throughout the entire lifecycle of a transaction, according to Amex, starting from the initial product search and purchase authorization all the way through to post-sale processing and dispute management.
Dubbed the "Intent Intelligence Framework," the company hopes that its latest creation will eventually become an industry-wide standard for all AI-powered transactions. As artificial intelligence continues to reshape how consumers interact with merchants, the financial ecosystem is bracing for a profound transformation in digital retail.
"These new transactions will have a consumer, a merchant, and an agent," explained Hannah Bozian, Amex’s Vice President of Agentic Partnerships and Strategy, during an interview with TechCrunch. She emphasized that the credit card giant wanted to ensure that agentic transactions remain as seamless, trusted, and secure as card members have historically expected them to be.
The release of this new framework directly follows the company’s April launch of Amex’s Agentic Commerce Experiences, commonly referred to as the ACE Developer Toolkit. That developer toolkit includes a growing suite of specialized services, which now officially incorporates the newly unveiled intent framework.
Amex noted that while it has not yet experienced a measurable increase in consumer disputes specifically driven by agentic commerce errors, it elected to build this technology proactively as this unique form of retail scales up globally. This proactive posture is largely driven by clear signals from consumers, who have demonstrated a strong willingness to embrace AI-driven shopping and research tools.
According to a recent Amex Trendex report, 66% of surveyed U.S. consumers indicated that they are likely to use AI agents over the next 12 months to research the best possible prices for products or services. Furthermore, 61% of respondents said they are likely to rely on AI agents for personalized product or service recommendations during the same timeframe, highlighting a rapidly growing consumer appetite for automated retail assistance.
To operationalize this vision, Amex’s new intent framework kicks off with a specialized Create Intent API. This application programming interface is designed to capture the customer’s exact purchase instructions alongside any additional constraints they may have included in their original prompt, such as the preferred merchant’s name, a strict maximum budget, and specific product preferences. While the exact method by which this intent is collected may vary depending on the underlying AI model involved in the interaction, the overarching plan is to build a framework flexible enough to support all existing and emerging agentic commerce models, Bozian explained.
Once the agent compiles the parameters, the customer must authenticate their identity and formally sign off on the intent captured by the agent. Following this sign-off, an Intent ID—which serves as a unique identifier representing the precise intent—is generated alongside a Proof of Intent Token. This token securely encapsulates the customer’s high-level constraints and parameters.
The token is subsequently utilized to obtain a secure digital payment credential that the AI agent will present at checkout, rather than exposing the customer’s actual credit card number to the merchant or the bot’s internal processing systems. By adopting this architecture, Amex states that purchases will remain strictly secure while staying thoroughly aligned with what the customer originally authorized the agent to perform.
In addition to securing the checkout process, the Intent ID and accompanying token can be seamlessly leveraged across various other stages of the transaction lifecycle. This infrastructure helps companies effectively manage complex operational areas such as fraud prevention and dispute resolution down the line, according to Amex.
"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 practical value of this approach, she offered a hypothetical scenario: a customer instructs an AI agent to purchase a pair of white Nike sneakers in a size 9, but the agent mistakenly submits a request to purchase a completely different pair of shoes from Adidas. In such a scenario, Amex’s system could intercept the error and prevent the transaction from going through entirely because the agent deviated from the explicit parameters of the user’s instructions.
Bozian also pointed out that while many consumer-facing agents today are programmed to strictly ask for manual customer consent before executing any financial purchase, Amex foresees a near future where agents will be trusted to act with significantly higher degrees of autonomy in specific, routine contexts. For example, a consumer might grant standing permission for their AI agent to automatically reorder household staples such as paper towels or toilet paper—product categories where the only truly meaningful constraints are the price threshold and the package size, rather than a specific brand loyalty or merchant preference. In these automated scenarios, having a dedicated security technology constantly in the loop performing extra validation checks ensures that the executed transactions consistently match the true intent of the consumer.
Maintaining a transparent, verifiable log of what was authorized, by which specific agent, and at what exact time could also provide an invaluable mechanism for addressing fraudulent activities carried out on behalf of or through compromised agents, she noted.
American Express is already actively engaged in discussions with various commercial partners across the broader payments ecosystem regarding the implementation of ACE. This rollout includes the specialized agentic tokens housing Amex’s proprietary payment credentials, as well as the various supporting components included within the developer kit.
The company stated that it intends to continue developing and expanding the ACE ecosystem aggressively as the agentic commerce landscape matures and evolves in the coming years.
Industry observers and fintech professionals will get a closer look at these developments later this year. On October 13 at TechCrunch Disrupt 2026’s Industry Stage, Amex VP of Agentic Partnerships and Strategy Hannah Bozian is scheduled to discuss the company’s new "Intent Intelligence Framework" in much greater detail, shedding further light on what this technological leap means for the broader world of agentic commerce and digital payments.
