Six banks propose five principles for purchases made by AI agents

Six international banks published a set of five principles on Tuesday, September 22, 2026, to guide purchases and payments made with the help of artificial intelligence agents. ASB Bank, Bank of America, Capital One, Commonwealth Bank of Australia, ING Group and NatWest Group identify transparency, safety, privacy and data, choice and interoperability as the foundations of agentic commerce.

The paper is not a regulation, imposes no conduct and sets no timetable. Its authors explicitly describe the principles as voluntary and non-binding. The proposal is a joint industry position for a stage in which digital assistants move beyond recommending products and begin entering information, authorizing payments or completing purchases for users.

From recommendation to autonomous purchase

Agentic commerce covers several levels of autonomy. In one scenario, an agent searches for options and a person executes the purchase. In another, the user pre-approves limits and conditions so the system can complete the transaction later. At the far end, an agent identifies a need and buys without a final review.

This shift changes the traditional e-commerce chain. In addition to the consumer, merchant, bank and payment network, agent providers, digital wallets and technology platforms become participants. The core problem is no longer limited to protecting card details. The system must also prove who authorized the action, what limits were granted, which company influenced the recommendation and who should resolve a dispute.

The banks say consumers worry about incorrect purchases, overspending, fraud and uncertainty over whom to contact when something goes wrong. Merchants may also face more chargebacks or lose their direct connection with customers. These are concerns presented by the authors, not evidence that every shopping agent produces those outcomes.

What the five principles would mean in practice

Transparency begins with identification. Consumers, merchants and intermediaries should know when an agent is involved, whom it represents and how it prioritizes products or payment methods. That includes disclosing sponsored options and conflicts of interest, such as a recommendation that earns the provider a higher commission instead of delivering the best value to the buyer.

Under the safety principle, users should be able to inspect and manage the authority delegated to an agent. Credentials and purchase intent should move through secure, auditable methods, with authentication available when a party exposed to liability considers it necessary. Dispute processes should also involve all responsible participants instead of leaving a consumer and merchant to reconstruct an automated decision on their own.

Privacy and data form a separate principle because these transactions create new information: conversations, preferences, spending limits, decision logs and purchase details. The banks call for collecting only the data required for each function, obtaining consent for additional uses and maintaining enough records to prove instructions, authentication and outcomes. The challenge is to balance two needs: reducing data exposure while preserving evidence for fraud investigations.

Choice and interoperability are meant to stop a platform from forcing people to use its wallet, payment method or agent. Consumers and merchants should be able to choose services without arbitrary restrictions, while different systems should exchange essential information securely.

The proposal does not yet settle liability or enforcement

The paper sets a direction but leaves difficult decisions for the next stage. It does not determine who pays when an agent exceeds its authority, how to prove that consumer intent was interpreted correctly or which regulator would oversee the participants. It also does not define a shared technical standard for identity, consent and revocation.

The consortium says a later paper will address implementation through protocols, standards and policies. Until then, the five principles should be understood as a banking-industry statement, not as protection already available to shoppers. The paper’s clearest contribution is separating convenience from trust: automating the purchase click is relatively simple; preserving control, dispute rights and freedom of choice requires the entire chain to record and explain what the agent did.