
American Express adds purchase-intent authentication to its AI commerce toolkit
The AMW Read
An incremental addition to Amex's existing commerce toolkit could shape payment authorization for AI agents, but partner implementation and effectiveness are not established.
American Express adds purchase-intent authentication to its AI commerce toolkit
American Express has introduced an Intent Intelligence Framework to capture and authenticate customer instructions before an AI agent completes a purchase. It is part of the Agentic Commerce Experiences (ACE) Developer Toolkit, which debuted in April. A Create Intent API records constraints such as merchant, budget and product preferences. After the customer authenticates and approves, the framework generates an Intent ID and Proof of Intent Token, enabling the agent to obtain a secure payment credential instead of the actual card number.
The development puts agentic commerce at the intersection of autonomous software and financial-services infrastructure. An agent's ability to execute a purchase creates a separate requirement: establishing whether that purchase matches the customer's authorization. Amex brings that check earlier into the transaction flow, with identifiers that can also support fraud prevention and dispute resolution. This gives payment networks a potential role in governing agent actions through the payment workflow. Amex hopes its approach becomes an industry standard, but the report establishes a framework launch, not broad adoption or demonstrated reductions in unauthorized transactions.
For builders, the concrete integration question is how to carry approved merchant, budget and product constraints from the customer's instructions into payment authorization. Evaluating ACE should include whether a purchase outside those constraints is blocked before completion and whether its intent record supports a subsequent dispute. Amex says it has not seen a rise in agentic-commerce disputes and is discussing ACE implementation with payments partners; adoption and effectiveness remain the next evidence to assess.