Key takeaway: The paper released by the banks is a voluntary, nonbinding principles paper. It sets no implementation timetable, so nothing changes yet. Even so, it signals where major issuers want liability rules to land. Merchants should watch for the banks’ promised follow-up paper on turning these principles into actual protocols and industry standards. That is where enforceable dispute and liability mechanics would take shape.
Six major banks are calling for AI shopping transactions to carry an auditable record. That record would span what a customer asked an agent to do, what authority they granted, and what happened before and after payment.
ASB Bank, Bank of America, Capital One, Commonwealth Bank of Australia, ING, and NatWest laid out the proposal in a paper published September 22, 2026, titled “Building Trust in Agentic Commerce.” The banks said providers should preserve evidence of consumer instructions, authentication, intent, transaction decisions, and outcomes, including any warnings or interventions along the way. That record would give participants a basis to investigate scams, recover money, and resolve disputes.
The paper defines agentic commerce broadly. On one end is an agent that simply helps a person search for a product before checkout. On the other is one that selects and purchases an item autonomously after an initial instruction, without the customer reviewing the final choice.
The banks flagged that issuers and acquirers may lack real-time access to an agent’s identity, the merchant of record, or the customer’s actual intent and purchase details. That gap becomes a problem when something goes wrong.
“Merchants fear disputes and chargebacks arising from decisions they did not control,” the paper said. To address that, the banks proposed that liability should reflect where an error or risk actually entered the transaction. They also called for dispute processes that involve the relevant participants, and for secure, auditable methods of authorizing agent purchases and payments.
Why it matters: AI agents take on more of the buying decision. The industry’s existing dispute machinery was built around a human clicking buy. It doesn’t cleanly answer who’s responsible when an agent overspends, buys the wrong item, or gets scammed. This proposal is notable because it comes from issuers, not merchants. By tying liability to where the error actually occurred rather than treating the merchant as the default bearer of loss, it could reshape how chargeback responsibility gets allocated as agentic commerce scales.
Source: PYMNTS
