Imagine having a personal assistant who checks your bank balance, compares savings rates, pays bills and moves money to get you a better deal. Now imagine that assistant is an AI agent operating through your online banking account. Convenient? Absolutely. But what happens when it does something you never asked it to do?
That is the concern facing some major US banks.
AI agents have entered banks. Well, at least some customer accounts. Willingly, in broad daylight, with due permission of their respective customer.
According to Reuters, banks including JPMorgan Chase, Bank of America and Morgan Stanley are restricting third-party AI agents from accessing customer accounts. The report, which draws on people familiar with the situation, describes concerns about agents attempting actions that appeared to go beyond users’ instructions.
Importantly, this does not necessarily mean hackers have broken into these accounts.
In some cases, customers have voluntarily shared their banking usernames and passwords with AI tools. The problem is that banks may struggle to distinguish legitimate, user-directed activity from an agent acting unpredictably. Morgan Stanley has said it had not observed rogue AI agents accessing its platforms.
So why are banks worried?
Unlike a chatbot that simply answers questions, an AI agent can carry out tasks. Think of the difference between asking someone how to pay a bill and handing them your wallet, bank card and permission to spend.
An agent with broad access could expose sensitive financial information, initiate unintended transactions or move money incorrectly. If criminals exploit such access, the consequences could be even worse.
Could one AI agent put other bank customers at risk?
The risk could extend beyond the person who gives an AI agent access to their account. Imagine an agent that has been given the keys to one apartment block. It should only enter the apartment of the person who hired it. But if the building’s access system has a flaw, the same key might expose other apartments.
A similar problem could arise in banking. If an AI company suffers a data breach, attackers might steal stored banking credentials or exploit weaknesses in its software. If the agent has access to several accounts, or a flaw allows it to cross boundaries between customers, the damage could spread. A compromised agent might also send fraudulent payment instructions or expose financial information.
These are hypothetical scenarios, not evidence that AI agents have already compromised large numbers of unrelated bank accounts. The danger depends on how the agent is built, what permissions it receives and how well the bank separates customers’ data. The US banking regulator’s guidance on third-party technology risks highlights why banks must monitor external providers and maintain strong safeguards.
Can banks legally ban customers from using AI agents?
In the US, banks can generally restrict or block third-party AI agents from accessing their online banking systems, subject to their contracts and applicable laws. This is different from making it illegal for customers to use AI altogether.
The distinction is becoming clearer in banks’ own rules. JPMorgan Chase’s terms say an AI agent must identify itself and comply with the bank’s conditions, and the bank can restrict or terminate its access.
In practice, a bank could block an automated tool, limit it to reading account information or require the customer to use an approved connection. If a customer repeatedly violates account terms or creates a security risk, the bank may also have grounds to restrict digital services or, where permitted by the agreement and law, close the account.
But this does not give banks unlimited power. Consumer-protection laws and rules governing unauthorised transactions may still apply. Whether a bank can close an account, refuse a transaction or hold a customer responsible for an AI agent’s mistake depends on the circumstances and the applicable law.
But the technology could also benefit customers.
As McKinsey explains, financial AI agents could compare banking products, identify better interest rates and help customers manage their money more efficiently. They could become digital financial assistants that work around the clock.
The answer, therefore, is not necessarily to ban AI agents forever. It is to give them carefully defined permissions. An agent that can read transactions but cannot transfer money is much less risky than one with unrestricted access. Human approval for large payments, spending limits and clear activity logs could provide additional safeguards.
Ultimately, the issue is not whether people should be allowed to use AI. It is who controls the technology when real money is involved. Banks will need ways to verify an agent’s identity, limit its permissions and stop suspicious activity before it causes harm. Customers, meanwhile, should not have to surrender control of their money simply to benefit from automation.
For now, the message is simple: AI agents could make banking smarter, but giving them your banking password is a serious gamble. Until banks offer secure, officially supported ways for these agents to operate, convenience should not come at the cost of control over your money.
Reference:
- Reuters — US banks are blocking third-party AI agents’ access to customer accounts (October 22, 2026).
- Axios — Unvetted AI agents are accessing US bank accounts (October 22, 2026).
- McKinsey — Agentic AI in retail banking: The customer effect (explores potential benefits and disruption to banking).