Individual AI agents are increasingly conducting transactions on behalf of users, but given the uncertainty of large language models, AI robots are also prone to doing things they shouldn't. In such cases, this means that AI errors may occur and unauthorized transactions may be initiated.
To address this issue in advance, American Express has introduced a new framework aimed at enabling AI agents to capture and verify customers' purchasing intentions before completing transactions.
American Express stated that the core of this concept is that there should be a way to ensure that customers' purchase instructions are retained throughout the entire transaction lifecycle, from the purchase itself to all subsequent stages.
This framework has been named “Intent Intelligence Framework (Intention Intelligence Framework)”. The company hopes that it will become the industry standard for driving transactions with AI.
"These new types of transactions will involve consumers, merchants, and agents," said Hannah Bozian, Vice President of Agency Cooperation and Strategy at American Express, in an interview with TechCrunch. She stated that the company aims to ensure that agency transactions remain "seamless, trustworthy, and secure, just as our cardholders expect them to be."
The release of this framework follows closely on the company's launch of Amex Agentic Commerce Experiences (Proxy-based Business Experience) in April of this year, which was followed by ACE Developer Toolkit ( ACE Developer Toolkit). This toolkit, which includes a range of services, has now also been incorporated into this framework.
American Express stated that they have not yet seen an increase in disputes resulting from agency-based commerce, but as this business model expands, the company hopes to proactively develop this technology. Part of the reason is that consumers have shown a willingness to try agency-based commerce.
The company stated to TechCrunch that in the most recent Amex Trendex report, 66% of American consumers surveyed indicated they might use AI agents to research the best prices for certain products or services; 61% of respondents said they might use AI agents to obtain product or service recommendations within the next 12 months.
Firstly, American Express's new intent framework begins with a Create Intent API. This interface will capture customers' purchase instructions, as well as any additional constraints they may include in their requests, such as merchant names, maximum budgets, and other product preferences. Bozian indicates that the method of capturing intent may vary depending on the AI model involved, but the company's plan is to make this framework flexible enough to support all agency-based business models.
Subsequently, the customer will verify their identity and confirm the intended content captured by the agent.
Next, the system will create a Intent ID, which is the unique identifier representing that intent. At the same time, it will generate Proof, of, and Intent Token (intent proof tokens), which contain the customer's high-level constraints. These tokens will be used to obtain a secure digital payment credential. The agent will use this credential during checkout, rather than the customer's actual credit card number. American Express states that this not only ensures the security of the purchase process but also guarantees that the transaction is consistent with what the customer has authorized the agent to perform.
American Express stated that this Intent ID and tokens can also be used in different stages of the transaction process to assist companies in areas such as fraud prevention and dispute resolution.
Bozian indicates: "The difference between our approach and some other implementation methods is that we introduce intent at an earlier stage of the transaction process. This is very important because it helps to reduce certain risks before they occur."
She gave an example: a client might request that an agent purchase a pair of white Nike sneakers size 9, but the agent submitted a request to buy Adidas shoes instead. In such a case, American Express can prevent the transaction from going through, as the agent made a mistake.
Bozian also points out that although many agents now seek customer consent before making purchases, American Express anticipates that in the future, under certain circumstances, agents may have much more autonomy. For example, consumers might allow agents to automatically restock household items such as tissues or toilet paper—where the only truly important constraints are likely to be price and specifications, rather than the merchant or brand. In such scenarios, it makes sense to use technology to perform additional checks within the process to ensure that transactions align with the customer's intentions.
She also stated that keeping logs of "which agents obtained what authorizations at what times" can also help in dealing with fraud issues caused by the agents themselves.
American Express stated that the company is already discussing with partners in the payment ecosystem how to implement ACE, which includes proxy tokens containing American Express payment credentials, as well as other components of the developer toolkit.
The company stated that with the development of the agency-based business ecosystem, it plans to continue to actively promote the subsequent development of ACE.
On October 13th, at the Industry Stage event of TechCrunch Disrupt 2026, Hannah Bozian, Vice President of Agency Cooperation and Strategy at American Express, will introduce in more detail the company's new “Intent Intelligence Framework” and what it means for the agency business world.
