American Express is launching a new system to prevent AI agents from making purchases the customer didn’t explicitly authorize. The move addresses growing concerns about AI tools–and the large language models behind them–making transactions that deviate from a user’s intent. The company’s solution aims to ensure purchase instructions are captured and enforced throughout the entire transaction lifecycle.
From Research to Reality: Intent Intelligence Framework
The core of the initiative is dubbed the “Intent Intelligence Framework.” It kicks off with a “Create Intent API” that records what the customer wants—merchant, budget, product preferences, plus any other constraints. After capturing those instructions, customers authenticate their identity and explicitly approve the captured intent. Only then is an “Intent ID” generated, paired with a “Proof of Intent Token” which encapsulates their constraints. An agent uses that token instead of the customer’s actual credit card at checkout to stay aligned with what was authorized.
This intent-and-token approach serves multiple safeguards. The token can be used later to help with fraud prevention and dispute resolution, while also being flexible enough to support varying levels of AI agent autonomy. For instance, if someone asks an agent to buy white Nike sneakers in size 9, but the agent attempts to order Adidas instead, the system could reject the transaction before it goes through.
Why Amex Is Building Ahead
While there haven’t yet been spikes in disputes tied to AI-driven commerce, American Express says they’re acting proactively. As agentic commerce—where AI agents act on behalf of customers—gains traction, the financial risks and regulatory concerns are also growing. New survey data from Amex’s own Trendex research shows that two out of three U.S. consumers expect to use AI agents to find better prices, and a similar share plan to lean on them for recommendations within the coming year.
The Intent Intelligence Framework is part of Amex’s larger Agentic Commerce Experiences (ACE) Developer Toolkit, introduced this past April. The toolkit already includes services that support AI agents; the new framework adds stronger checks throughout the purchase process. Amex is now working with partners across the payment ecosystem to integrate these systems, with an eye toward evolving the standards as agentic commerce develops.
At Disrupt 2026, Amex’s VP of Agentic Partnerships and Strategy is set to lay out more details of this framework, including what parties are needed in each transaction and how risk is managed before it escalates.
This is more than a protection mechanism—it’s a push toward defining best practices for AI-driven commerce. By pulling “intent” upstream in the purchase flow, Amex is betting it can prevent many common misfires before they happen.
Context & What to Watch
AI agents aren’t just science fiction anymore. From scheduling tasks to shopping, voice assistants and bots are getting ever more empowered to make decisions. But the inherent unpredictability of AI means even simple purchase requests can go wrong—leading to unauthorized charges, wrong items, or brands, etc. Payment networks, card issuers, and regulators are scrambling to catch up.
Amex’s move dovetails with broader industry efforts to establish guardrails for agentic commerce. Tokens and intent verification are emerging tools to balance convenience with control. If Amex succeeds, others in the fintech space—or even retailers—may follow.
Putting the control in customers’ hands is the bigger shift. As agents become more autonomous—restocking household staples, for example—the foundational question isn’t if they’ll act, but under what conditions. Amex’s framework offers a path toward ensuring those conditions match what users want, not what went awry.
