Personal AI agents are often sold to customers as a way to take the pain out of online shopping by helping you find the right product, the best possible deal, and even handling payment and shipping. While AI companies push agentic tools, some of the world’s biggest banks have expressed concerns about what it means for customers and financial institutions.
A new "principles paper" from a collaboration of six banking groups details potential issues that may arise as agentic AI tools become more widely adopted. The report comes from ASB Bank, Bank of America, Capital One, Commonwealth Bank of Australia, ING Group, and NatWest Group.
The collaboration notes how agentic commerce has the “potential for higher rates of scams, fraud, and disputes.” It explores how agentic tools may increase the number of cases of people accidentally buying fake items online, thereby leaving customers out of pocket. It also cites safety concerns about websites that the average user may now recognize as unsafe, but an agent wouldn’t spot.
The paper says, "Malicious actors may attempt new attack vectors for scams and fraud, including compromising or impersonating AI agents and merchants, and engaging in new forms of social engineering."
Banks are also concerned about where the blame would lie if an agent bought a product and you’re unhappy with it. For example, if your personal AI buys the wrong item, would you charge it back to your bank, or would the fault lie with the customer, or even the agent? For now, this remains unclear, and the group argues that agentic tools need to make themselves known to retailers and payment integrations to help address potential issues.
The paper says the collaboration aims to “establish a framework for a future where AI agents could play a greater role” in product purchasing, with the partnership hoping to encourage other stakeholders, such as AI firms, to join in on planning for the future.
This report doesn’t reference any AI brands by name, but it likely refers to tools like Meta’s Muse, as well as agentic options from Anthropic, Google, OpenAI, and other tech brands. The paper comes days after Amazon blocked Meta's new agentic AI assistant from shopping on its website, citing safety and privacy concerns.