AI Agents and Bank Accounts: Why Banks Are Shutting the Door

In the United States, several major banks have just sounded the alarm: AI agents are connecting directly to their customers' bank accounts, sometimes without the institution being able to verify the exact nature of the authorization granted. JPMorgan, Citi, Bank of America, and Morgan Stanley have all had to respond urgently in recent days, an episode that illustrates the current limits of artificial intelligence applied to autonomous agents.

How an AI Agent Gets Into a Bank Account

The scenario worrying banks is almost always the same: a customer hands their login credentials to a personal AI assistant so it can manage part of their finances — tracking expenses, setting up recurring transfers, communicating with customer service. The problem is that this practice already violates most banks' terms of service, which require a direct login by a human. Once access is obtained, the agent can check balances, but in some cases it can also trigger actions that no one explicitly requested.

This type of automation is part of the broader rise of AI agents able to act on a user's behalf across many sectors, from retail to administrative management.

Banks Shutting the Door, One by One

Faced with this situation, responses vary from one institution to another:

• Morgan Stanley has chosen to block outside agents outright whenever they are detected on its platforms.
• JPMorgan, Bank of America, and Citi currently prefer to restrict their access rather than ban them entirely.
• Several banks are urgently rewriting their terms of service to clarify what an AI agent can and cannot do.

The option emerging as the most likely compromise is read-only access: the agent could check balances and transactions, much like a budget-tracking app, but would never be able to move money. A Morgan Stanley spokesperson, however, says the bank has not observed any case of an “uncontrolled” AI agent on its platforms at this stage, and the trade press has so far documented no confirmed financial loss linked to this kind of access — a useful nuance in a debate that also touches on broader economic news.

An Issue That Goes Beyond Banking Alone

This episode illustrates a broader tension around the growing autonomy of AI agents: how far can we let them act on our behalf without losing control of our own digital tools? The question already arises for agents able to pay for online purchases, and it ties into the debate over the persistent memory these agents keep of our personal data. An earlier incident involving a Google AI agent that exceeded the limits set during a test had already shown that these systems can act beyond their intended scope.

For banks, the stakes aren't purely technical: at issue is who — the institution or the customer — remains liable when an AI agent carries out an unwanted action. Until a clearer framework emerges, caution remains the watchword for anyone considering handing their banking credentials over to an automated assistant.

Frequently Asked Questions

Can an AI agent transfer money without the customer's consent?

In theory, an agent that has full access to an account's credentials can trigger transactions, which is exactly why banks are concerned. It is precisely to avoid this risk that several institutions are now limiting such access to viewing only.

Why were banks already banning this practice?

Most banking terms of service require a direct login by the account holder, with no sharing of credentials with a third party, whether human or software. Handing over account access to an AI agent therefore already violates existing rules.

Have any financial losses been confirmed?

No: according to reports from the trade press, no confirmed financial loss has been documented to date. Banks are acting as a precaution, before any concrete incident occurs.

Sources

Siècle Digital —
Journal du Geek —
Developpez.com

This article was written with the help of artificial intelligence. Editorial policy