Six Banks Publish AI Commerce Principles, Citing Liability Gaps

Six banks have published principles for AI agents that shop and pay for customers, seeking safer transactions and customer control while warning that failed purchases can leave liability unclear. The paper identifies risks from overspending and incorrect orders to scams, fraud and agent impersonation.

Bank of America, Capital One, ASB Bank, Commonwealth Bank of Australia, ING Group and NatWest Group published Building Trust in Agentic Commerce. The banks say the framework is meant to support responsible agentic commerce while preserving customers’ and merchants’ choice, control and flexibility.

A framework for the handoff from shopper to agent

The paper proposes five principles: transparency, safety, privacy and data, choice, and interoperability. Its focus is not just whether an agent can find a product, but what happens when it takes action for a customer and exceeds the authority that customer intended to give it.

That handoff can create competing expectations among consumers, agents and merchants about the product or service to be delivered, as well as when and how it should arrive. The banks also warn that actions by shopping agents could raise merchant disputes and credit-card chargebacks.

50%Used AI help for a retail purchase

PYMNTS Intelligence reported that 50% of Americans had made a retail purchase with AI assistance.

24%Willing to let an AI shop and pay

The same PYMNTS Intelligence research found that 24% were willing to let an AI agent shop and pay.

That difference separates AI assistance from a system carrying a purchase through payment. The banks’ principles put customer and merchant control alongside transaction safety, reflecting the added stakes once an agent acts rather than simply helps a shopper research options.

The risks range from mistakes to manipulation

• An agent could buy the wrong item, spend too much, or lose money to a scam or fraud.
• Bad actors could compromise or impersonate agents and merchants, including through new social-engineering tactics.
• Providers could handle card details unsafely or prioritize payment methods with weaker protections.

The difficult question comes after a disputed purchase. The banks say liability allocation can be unclear when agentic-commerce transactions go wrong, and existing dispute processes may not involve every relevant party in the value chain. Those gaps could complicate the path from a customer complaint to a decision about responsibility.