News

Six Banks Publish Principles for Trusted Agentic Commerce, Including Liability

Agentic Commerce Atlas · · Updated

Good for: Leaders · Builders

On 23 September 2026 six banks on three continents, Commonwealth Bank of Australia, ASB Bank, Bank of America, Capital One, ING Group and NatWest Group, published Building Trust in Agentic Commerce, a set of voluntary principles for AI agents that buy on behalf of consumers and merchants. The principles cover five areas, transparency, safety, privacy and data, choice, and interoperability, and say that when an agent purchase goes wrong, liability outcomes should reflect where risks and errors are introduced. A second paper on how to implement them is planned.

What the six banks’ principles say

The paper sets one headline principle for each area, with numbered sub-principles beneath it.

  • Transparency. “AI agents should be transparent with the consumers and merchants they serve.” All parties should know when an agent is involved in a transaction and on whose behalf it acts, and consumers and merchants should be able to see how agents rank options, including sponsored ones.
  • Safety. “Consumers and merchants should be able to transact safely in agentic commerce.” Consumers should control the authority they grant agents, including viewing and managing delegations. Payment credentials should be entered, and purchase intents and payments authorized, by secure and auditable methods, with any party holding a potential liability able to require authentication.
  • Privacy and data. Providers should keep auditable records of consumer instructions, authentication, intent and transaction outcomes, access only the data they need, and use or share data further only with consent.
  • Choice. “Consumers and merchants should be able to choose which agentic commerce services they use.” Service providers keep the choice of which agentic services they support on safety, commercial and legal grounds, and the merchant of record should be clear.
  • Interoperability. Industry should build interoperability for the core functions behind the principles while leaving room to compete on value-added services.

How the paper treats liability for agent purchases

Liability sits in the safety section. Among the risks the paper lists are mismatched expectations about “who would be liable if an AI agent exceeds its authority” and that “there is unclear and inefficient allocation of liability,” with dispute processes that “do not involve all relevant parties across the value chain.” Principle 2.5 responds: “Dispute resolution processes should involve all relevant parties to agentic commerce transactions and be fair and efficient, with liability outcomes reflecting where risks and errors are introduced.” In the announcement, Bank of America’s Mark Monaco names “liability management and customer protection” among areas that will need “thoughtful approaches.”

The paper also flags that issuers and acquirers may lack real-time access to an agent’s identity, and that agents and merchants can be impersonated.

Why the bank principles matter for agent payments

This is a group of card issuers and account providers stating what they expect. Several of the principles cover delegations a consumer can view and revoke, authentication that the party bearing the risk can demand, and auditable records of intent, the territory of payment authorization mandates and tokens. The paper names no specific protocol or network program, though it says the follow-up paper will cover implementation through protocols.

The six banks are based in Australia, New Zealand, the US, the UK and the Netherlands.

What the principles paper does not do

The paper states that “the principles described are voluntary and non-binding” and that nothing in it requires the authors or anyone else to take any particular action. It does not set liability rules, name a dispute process, or commit the banks to a standard. The banks say they will “detail in a subsequent paper how the principles can be implemented,” and invite engagement from customers, industry and other stakeholders. No date for the second paper has been given.

See also agentic commerce risks and readiness, AI shopping agent payment authorization, and the agentic commerce timeline.

Frequently asked questions

Which banks wrote the agentic commerce principles paper?

Six banks: Commonwealth Bank of Australia, ASB Bank, Bank of America, Capital One, ING Group and NatWest Group. The paper is titled Building Trust in Agentic Commerce and was announced on 23 September 2026.

What does the paper say about liability for agent purchases?

It lists unclear and inefficient allocation of liability as a safety risk, including who is liable if an AI agent exceeds its authority. Principle 2.5 says dispute resolution should involve all relevant parties to agentic commerce transactions and be fair and efficient, with liability outcomes reflecting where risks and errors are introduced.

Are the principles binding?

No. The paper says the principles are voluntary and non-binding and do not require the authors or other industry participants to take any particular action. The banks say a subsequent paper will detail how the principles can be implemented.

Primary sources

  1. Global banks collaborate on principles for trusted agentic commerce · Commonwealth Bank of Australia, 2026-09-23
  2. Building Trust in Agentic Commerce: Customer-centric principles for a trusted and thriving ecosystem · ASB Bank, Bank of America, Capital One, Commonwealth Bank of Australia, ING Group and NatWest Group, 2026-09-23