As personal artificial intelligence agents increasingly step into the role of financial proxies—browsing the web, comparing prices, and executing purchases on behalf of their users—the payments industry is facing an unprecedented technical challenge. Because large language models are inherently nondeterministic, AI bots carry a distinct risk of misfiring, misunderstanding complex prompts, or executing unauthorized transactions that deviate entirely from what their human users originally intended.
In an effort to get ahead of this emerging vulnerability before it transforms into a widespread crisis, financial services giant American Express is introducing a robust new framework designed specifically to capture, verify, and authenticate a customer’s true purchase intent before an AI agent is ever permitted to complete a financial transaction.
The core philosophy behind the initiative is to establish an unbroken chain of custody for a consumer’s instructions, ensuring that purchase parameters are reliably carried throughout the entire transaction lifecycle—from the initial product search to the checkout counter and everything that follows. Dubbed the “Intent Intelligence Framework,” American Express hopes the new architecture will eventually serve as a dependable industry standard for the rapidly expanding ecosystem of AI-powered commerce.
The rollout of this framework builds directly upon the company’s previous efforts in the space, following the April launch of Amex’s Agentic Commerce Experiences, or the ACE Developer Toolkit. That initial toolkit provided developers with a foundational suite of services, which has now been significantly bolstered by the addition of the new intent-capturing framework.
According to American Express leadership, the company has not yet recorded a measurable spike in formal consumer disputes stemming directly from agentic commerce. However, the financial institution is deliberately taking a proactive stance as this novel form of digital retail begins to scale up. That strategic foresight is fueled in large part by clear signals from consumers, who have demonstrated a strong appetite for delegating shopping tasks to autonomous software.
Data from a recent Amex Trendex report highlights this consumer shift, revealing that 66 percent of surveyed U.S. consumers are likely to utilize AI agents to research the absolute best price for a product or service over the coming year. Furthermore, 61 percent indicated they are likely to lean on AI agents to secure product or service recommendations during the same timeframe, demonstrating that consumer trust in automated assistance is rapidly growing.
To address the inherent risks of this shift, the American Express intent framework kicks off with a dedicated Create Intent API. This application programming interface is engineered to capture a customer’s precise purchase instructions alongside any specific constraints included in their original request—such as a designated merchant’s name, a strict maximum budget, and various other granular product preferences. While the exact method of collecting this intent may vary depending on the underlying AI model utilized by the consumer, the overarching design of the framework is intentionally flexible, ensuring it can seamlessly support a wide variety of agentic commerce models currently entering the market.
Once the agent compiles the parameters, the customer must personally authenticate their identity and explicitly sign off on the intent captured by the system. Following this validation, the system generates an Intent ID—a unique digital identifier representing the specific transaction intent—alongside a Proof of Intent Token that encapsulates the customer’s high-level financial and logistical constraints.
Crucially, this specialized token is subsequently used to obtain a secure digital payment credential. Instead of exposing the customer’s actual credit card number to the AI agent or the merchant at checkout, the agent uses this restricted tokenized credential. By handling the transaction this way, American Express ensures that the purchase remains strictly aligned with what the customer originally authorized, maintaining a high standard of security.
Furthermore, this unique Intent ID and its associated token can be leveraged across various stages of the transaction lifecycle to help companies effectively manage critical operational areas such as fraud prevention and dispute resolution.
By capturing intent at the very beginning of the transaction flow rather than at the end, American Express aims to mitigate risk before it even materializes. Hannah Bozian, the Vice President of Agentic Partnerships and Strategy at American Express, explained the practical mechanics of this preventative approach in an interview, noting that these new transactions fundamentally involve three distinct parties: a consumer, a merchant, and an agent. The primary goal, she emphasized, is to ensure that agentic transactions remain as seamless, trusted, and secure as card members have historically expected.
To illustrate the practical value of early intent capture, Bozian offered a clear hypothetical scenario. Imagine a customer explicitly instructs their personal AI agent to purchase a specific pair of white Nike sneakers in a size 9. However, due to a hallucination or an algorithmic misinterpretation, the AI agent instead submits a purchase request for a completely different brand of shoes from Adidas. In a traditional automated checkout setup, that error might go unnoticed until the package arrives at the customer’s door. With the Intent Intelligence Framework, American Express possesses the capability to intercept and prevent the transaction from processing entirely, recognizing that the agent’s final action directly violates the customer’s established intent.
Looking toward the future of retail, Bozian also pointed out that while the vast majority of AI agents today still pause to explicitly ask for customer consent before executing a final purchase, the industry is steadily moving toward a reality where agents will act with significantly greater autonomy. For example, a consumer might grant their AI agent ongoing permission to automatically re-order routine 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 loyalties or merchant restrictions. In those highly automated scenarios, having a dedicated technological safeguard in the loop to perform continuous validation checks becomes essential to guarantee that every single transaction precisely matches the user’s overarching intent.
Maintaining a transparent, immutable log detailing exactly what was authorized, by which specific agent, and at what precise time will also provide a vital mechanism for combating fraud perpetrated on behalf of automated agents.
As the underlying technology matures, American Express is already actively engaged in discussions with various industry partners across the broader payments ecosystem to facilitate the implementation of the ACE Developer Toolkit. This includes integrating the specialized agentic tokens that house secure Amex payment credentials alongside the broader components of the developer kit. The company has confirmed its commitment to actively developing and refining the ACE ecosystem as the broader landscape of agentic commerce continues to shift and expand.
Industry observers and fintech professionals will get a closer look at these developments later this year, when Hannah Bozian takes the stage at TechCrunch Disrupt to discuss the finer details of the Intent Intelligence Framework and its broader implications for the future of digital payments.
