AI Agents Are Logging Into Bank Accounts And Banks Can’t Tell

Major banks are confronting a problem with AI agents that will shape the future of agentic commerce. They cannot always tell whether a customer or an AI agent is acting inside an account. An AI agent is software that can work toward a goal by taking steps, using tools or accessing services, and responding to what it encounters along the way.

The scale is growing quickly. IDC projects that active AI agents will increase from 31 million in 2025 to 2.45 billion by 2030.

Banks including JPMorgan Chase, Bank of America and Wells Fargo are working to contain risks from third party AI agents, Axios reported. In some cases, agents appeared to take actions independent of a user’s instructions. Banks are restricting access as they assess the risks.

Some customers have shared their banking credentials, such as usernames and passwords, with AI agents so they can handle financial tasks. That can make it difficult for banks to tell when a customer is acting and when an agent is acting on the customer’s behalf.

The same question is emerging in government. Will Americans trust AI agents to verify their identity and act on their behalf?

AI Agents Expose The Limits of Shared Credentials

When an agent enters a bank account using a customer’s credentials, the bank sees the customer’s identity. It may have no reliable way to know whether the customer is checking a balance, an agent is following a request, or the agent has gone beyond what the customer intended.

That creates a basic control problem and governance issue. What was the agent allowed to do? Who gave it permission? Could it move money, change account details or contact the bank? Does that permission expire? Can the customer revoke it immediately?

A shared password answers none of those questions. It gives the agent access while obscuring its identity and authority.

A Password Cannot Define An AI Agent’s Authority

Agentic Commerce Needs Accountable Access For AI Agents

There are hard questions to resolve.