Invest1 publisher3 min readPublished
Six banks propose voluntary principles for AI shopping agents to declare who they work for at payment
NatWest, Bank of America, Capital One, ING, Commonwealth Bank of Australia and ASB published five voluntary principles on Sept. 22. The firms running agents today did not sign it, and all six co-authors are banks.
The Investor · Invest desk

What happened
- Six banks published joint principles on Sept. 22 proposing that all parties to an agentic commerce purchase know when an AI agent is involved in the transaction and who that agent is acting for.
- The paper, Building Trust in Agentic Commerce, was co-authored by NatWest Group, Bank of America, Capital One, ING Group, Commonwealth Bank of Australia and New Zealand's ASB Bank.
- Transparency leads the five principles and covers how agents rank options, including sponsored ones that pay their providers higher commissions even when they are not the best value for the consumer.
- The principles are voluntary, and the banks' stated next steps are a consultation with policymakers and a second paper on how to implement them.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- cost Putting liability where the risk or error was introduced would move dispute costs toward whoever built the agent, and off the merchants the paper says fear chargebacks arising for reasons outside their control.
- constraint A disclosure duty on sponsored ranking narrows what an agent provider can earn by routing a purchase to the product or the payment method that pays it the most.
- decision Every other bank and payments firm now has a public invitation on the table, and joining means accepting the disclosure and liability lines as the six banks have written them.
The commercial content of the paper is in two lines. One asks that an agent disclose how it ranks options, sponsored ones included [6]. The other is a finding: the authors say some providers ask consumers for their card details and then key them into websites, and some prioritise payment methods that offer less protection [14]. The paper's own explanation of why ranking matters is commercial. An agent may promote a product or a payment method because it gives its provider a larger cut, such as a higher commission, even when that is not the best value for the consumer [7].
Disclosure at the point of payment is a request for a data field. The banks want every party to a purchase to know that an agent is involved and who it is acting for [3]. When a provider types a card number into a checkout page, the authorisation that reaches the bank does not include that identification [21]. The safety principle asks that agents enter payment credentials in a secure, auditable manner, and that consumers be able to view and manage the delegations they have granted [8].
The dispute principle is the costliest one. The banks ask that every party take part in the dispute process and that liability sit where the risk or error was introduced [9]. The paper does not quantify the transfer. It says merchants fear disputes and chargebacks rising for reasons outside their control [15].
Enforcement is the open question. The principles are voluntary [4], all six co-authors are banks [19], and the six have invited other banks and payments firms to apply them and help build a blueprint for implementation [18]. Robinhood, which turned on credit card spending for AI agents in May while most finance companies were still in pilot mode, according to Cryptopolitan [17], is not one of the six [20]. Agentic commerce has moved from closed trials to live pilots and early scaling faster than formal governance, the banks said [16].
"Consumers are unclear if AI agents will act in their interests," the paper said [13]. "But for that potential to be realised, customers need to trust that they remain in control of how payments are made and that their money is safe," NatWest Chief Payments Officer Mark Brant said [12].
In my view transparency is the principle that gets diluted first, because it is the one that moves money: the safety and data lines cost the agent layer engineering work [8][10], while disclosing a commission changes who gets paid [7]. The other way this goes is that a large agent provider adopts an identification field because carrying the dispute cost is dearer than declaring itself, and the ranking disclosure travels with it. A named agent provider signing the ranking line, or a second paper that arrives with a field specification providers actually implement [4], would be evidence against me.
What to watch
- Whether the promised second paper contains a technical specification for an agent identification field or another set of voluntary principles.
- Whether any AI agent provider or card network signs up after the invitation to other banks and payments firms.
- Whether a policymaker in the consultation takes the disclosure line and turns it into a requirement.