Amex Unveils Framework to Prevent Unauthorized Purchases by AI Agents
American Express is introducing a new framework to ensure AI agents only make purchases users have authorized, potentially preventing costly errors and disputes as AI-driven shopping grows.
The last day to showcase your progress to over 10,000 tech leaders at Disrupt is on October 2nd. Book fair table now.
Disrupt the doors opening on October 13th. Get your card and bring someone with you with a 50% discount. Register now.
Personal AI agents are increasingly performing transactions on behalf of their users, but due to the non-deterministic nature of large language models, AI bots are prone to doing things they are not supposed to do. In this case, it increases the possibility of artificial intelligence error and making unauthorized transactions.
To overcome this problem, American Express is introducing a new framework designed to register and authenticate a customer's purchase intent before the transaction is completed by an AI agent.
The idea, Amex says, is that there should be a way to ensure that a customer's purchase instructions are followed throughout the transaction lifecycle, from the purchase to everything that follows.
Called the "Intelligent Purpose Framework," the company hopes to become an industry standard for AI-powered transactions.
"These new transactions will have a consumer, a merchant and an agent," explained Amex's vice president of partnerships and strategy, Hanna Bozian, in an interview with TechCrunch. He said the company wants to ensure that merchant transactions continue to be "unique, reliable and secure, as our card members expect."
The release of the framework follows the company's April launch of Amex's Agentic Commerce Experiences, or ACE developer toolkit, which includes a handful of services that now include the target framework.
Amex said it has yet to see an increase in disputes due to agency business, but wants to be proactive in building the technology because that type of business is large-scale. That's partly because consumers have indicated they're ready to try dealerships.
In a recent Amex Trendex report, 66 percent of U.S. consumers surveyed said they were likely to use AI agents to research the best price for a product or service, and 61 percent said they were likely to use it for a product or service recommendation in the next 12 months, the company told TechCrunch.
For starters, Amex's new intent framework starts with the Create Intent API, which captures the customer's purchase instructions along with other constraints they may have included in their request, such as merchant name, maximum budget, and other product preferences. Bozian said the way this goal is assembled may vary depending on the AI model involved, but the plan is for the framework to be flexible enough to support all agent business models.
The client then authenticates its identity and signs the target captured by the agent.
An Intent ID, a unique identifier representing the intent, is then created along with a Proof of Intent Token, which contains the client's high-level constraints. The token is used to obtain a secure digital payment credential that the agent uses when paying instead of the customer's actual credit card number. By doing so, Amex says, the purchase remains secure while still being consistent with what the customer authorized the agent to do.
This Intent ID and token can also be used in various parts of the transaction process, Amex said, helping companies manage areas such as fraud prevention and dispute resolution.
"Our approach differs from other implementations in that it brings intent earlier into the transaction flow. And that's important because it helps mitigate some of the risk before it happens," Bozian said. He gave an example where a customer asked an agent to buy white Nike sneakers in size 9, but the agent asked for Adidas shoes. In such a case, Amex can block the transaction because the agent made a mistake.
Bozian also notes that while many agents today ask for customer consent before making a purchase, Amex envisions a future in which agents may act too autonomously in some cases. For example, a consumer might allow their agent to reorder household staples like paper towels or toilet paper—categories where the only real constraints are price and size, not merchant or brand. In these scenarios, it makes sense for the technology in the loop to perform this additional check to ensure that the transactions match the customer's intent.
Having a log of what was authorized by which agent and when can also help address agent fraud, he noted.
Amex is currently in talks with partners across the payment ecosystem to implement ACE, which includes the agent tokens that hold Amex's payment credentials, as well as other parts of the developer kit.
The company said it plans to actively develop ACE as the operating business ecosystem evolves.
On Oct. 13 at the TechCrunch Disrupt 2026 Industry Stage, Hannah Bozian, Amex's VP of Agent Partnerships and Strategy at American Express, will discuss the company's new "Target Intelligence Framework" in more detail and what it means for the agent business world.
When you make a purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
Get 50% off a second pass The Disrupt experience is meant to be shared. Grab your pass and bring a colleague, partner or peer for 50% off. Cover more ground by building connections, building momentum, and discovering what's next in the startup ecosystem.
Elon Musk attacks Ambani over Jagmeet Singh's India Starlink launch delay
Batteries are already cheaper than the natural gas turbines used in many of Tim de Chant's data centers.
Remember Orkut? Its founder wants to bring back Jagmeet Singh
The United States barred Microsoft, Adobe and major IT companies from the green card program for skilled foreign workers Ayesha Malik.
Elon Musk questions Ambani influence as Starlink India launches on Jagmeet Singh stands
The 19-year-old founder of Cal AI just raised $10 million for his new AI startup Julie Burt.