The Maze: Six banks have published principles for agentic commerce: shopping in which AI software helps select or pay for a purchase. Bank of America, Capital One, ING, NatWest, ASB and Commonwealth Bank want safeguards to keep pace with the technology. The September 22 initiative brings together institutions from the US, Europe, Australia and New Zealand. Its immediate output is a shared position on trust and customer control; practical implementation remains ahead.
The shopper still needs a spending boundary. An assistant that recommends a product and one that completes a purchase create different risks. Consumers may worry about buying the wrong item, overspending or falling for a scam. The banks want the industry to address those gaps before broader adoption. For retailers, the useful question is whether an order carries enough evidence of what the customer actually authorized, especially when the customer never visits the checkout.
A faster payment can still produce a messy dispute. Entering a customer's card details directly into websites can create privacy and security concerns. Choosing a payment method may also change the protections attached to the purchase. Software can execute a transaction without resolving who should answer when it goes wrong. That leaves merchants with a practical dependency: payment partners and agent providers need a workable way to investigate errors together.
Five headings now need operational substance. Transparency, safety, privacy and data, choice, and interoperability form the common agenda. Interoperability means different services can work together. ING is inviting participation from merchants, technology providers and financial institutions to turn the principles into practical standards. The commercial test will be whether a retailer can accept automated demand without losing visibility into the purchase or having to support incompatible arrangements for every assistant.
The paper is voluntary, and the timetable is open. Building Trust in Agentic Commerce explicitly makes its principles non-binding. It does not prescribe a particular implementation method or deadline. A subsequent paper is intended to develop the framework. Calling this a new global rulebook would overstate what exists today: the banks have agreed on direction, while the details of applying it still need work. No conversion lift or adoption figure comes with the announcement.
Why it matters: AI-assisted sales will be more useful when merchants can tell a valid instruction from an expensive mistake. The banks' intervention puts authorization and accountability alongside convenience in the checkout conversation. Operators should watch the implementation work and how it handles disputes, rather than treating this announcement as a technical integration specification. A bot can finish an order faster than the industry can agree who owns the error.
Sources: PaymentsJournal | Bank of America | Principles paper | ING
