Why Some U.S. Banks Are Blocking AI Agents From Customer Accounts

An AI assistant that can check your bank balance or organize your finances sounds useful. But if it signs in with your banking password, your bank may see that as a problem—even when you gave the assistant permission.

That tension came into focus on October 22, when Axios reported that some major U.S. banks were detecting third-party AI agents inside customer accounts. Reuters also covered the report, while saying it could not immediately verify the findings. Some of the more serious details rely on people familiar with the situation, so it helps to separate what banks have said publicly from what has only been reported.

QUICK TAKE

• Some banks are limiting outside agents that sign in with customers’ banking credentials.
• The reports do not show that every named bank found rogue agents or customer losses.
• Check your bank’s rules and keep your banking password private.

An AI-themed illustration. Photo by Ant Rozetsky on Unsplash.

What the Reports Say—and What They Don’t

Axios reported that banks including Morgan Stanley, Bank of America, JPMorgan Chase and Wells Fargo were working to limit risks from outside agents. Sources told Axios that some banks had encountered agent activity they considered unauthorized. One concern was that an agent might contact a call center or take steps that did not match what its user intended.

That does not mean every named bank found a rogue agent, or that customer accounts were widely drained. The report did not give an incident count or a total for losses. Morgan Stanley told Axios it had not observed rogue AI agents accessing its platforms. Reuters said it could not immediately verify Axios’s report.

The careful reading is this: banks are concerned about how outside agents access accounts, and some restrictions have been reported. The public information does not establish that every bank has experienced the same problem.

What “AI Agent Access” Actually Means

A chatbot mainly responds to questions. An AI agent can also use tools: it may open a browser, fill in forms, send messages or attempt to sign in to a service. In banking, the way it connects matters.

Credential sharing: A customer gives an agent a username and password, and the agent signs in as if it were that customer. Banks may prohibit this even when the customer approves, because the login system cannot easily distinguish the customer from an automated tool.

An approved data connection: Depending on the bank and provider, this may let an app read balances or transactions without handing over a password. Read-only access is different from permission to transfer money or make payments. The available options vary, so customers should check directly with their bank.

A bank’s own assistant: This operates within the bank’s service and policies. These approaches can look similar from the outside, but they do not give an agent the same access or authority.

Digital banking on everyday devices. Photo by Neil Fernandez on Unsplash.

What Banks Have Said

JPMorgan Chase has said it does not allow outside agents to obtain customers’ banking credentials. Its chief executive, Jamie Dimon, has also raised concerns about letting unvetted agents into accounts. Charles Schwab’s chief executive has described plans for controlled connections that could let outside AI tools view information or help with research without moving money.

Other comments have been reported through news outlets rather than published as detailed bank policies. Axios’s sources say Morgan Stanley has started blocking third-party agents, while the bank’s spokesperson said it had not seen rogue agents. Those statements can both be true: a bank can restrict a type of access as a precaution.

Wells Fargo’s online banking notice says updated agreement terms are scheduled to take effect on November 24, 2026. The notice says third-party AI tools are not operated by the bank and that it may restrict or block certain tools. That future terms update does not, by itself, prove that Wells Fargo is currently blocking every AI agent. Read Wells Fargo’s notice.

The Chicago Federal Reserve Bank building. Illustrative photo by Joshua Woroniecki on Unsplash; this is not one of the banks named in the report.

Why Banks Are Cautious

Identity is hard to verify. A successful login shows that valid credentials were used; it does not prove who used them or whether an automated tool stayed within the customer’s instructions. If something goes wrong, that can make the activity difficult to investigate.

Control matters. Banks need to know which actions a customer approved, and when a person should confirm a payment or account change. A tool that can act quickly may create confusion if its permissions are unclear.

Responsibility is still unsettled. If a customer shares credentials with an outside tool, the customer, the bank and the tool provider may each have different terms and expectations. The answer can depend on the facts and the account agreement. Customers should contact their bank promptly if they notice activity they did not authorize.

Banks also have to consider security rules and fraud prevention. That does not mean every use of AI is unsafe. It means a bank may prefer a controlled, limited connection over an agent that signs in with a customer’s password.

What Customers Can Do Now

For now, the safest practical steps are straightforward:

• Do not give an AI tool your online banking username, password or one-time security code.
• Ask your bank whether it offers an approved connection for viewing balances or transactions.
• Read account notices and tool permissions before connecting a service.
• Turn on account alerts and review statements regularly.
• Pause before approving any payment, transfer or account change.

If an agent promises to manage your money automatically, check exactly what it can see and do. A read-only connection, where available, carries different permissions from a tool that can initiate transactions.

AI Tools Are Not All the Same

The restrictions being discussed are about how an outside agent connects to a bank, not a blanket verdict on every AI assistant. For example, Anthropic describes Claude Duet as an agent that pauses for approval before certain sensitive actions and does not directly see passwords or sign-in tokens. Those are Anthropic’s product claims; they do not change a bank’s own access rules.

We explained Claude Duet in our earlier guide.

What Remains Unclear

The public reports do not show how many customers have been affected, how often agents have acted outside instructions, or whether customers have lost money because of this activity. They also do not identify one AI product as responsible for the reported activity. Banks and AI companies are still working out how to offer useful features while limiting what an agent can do.

The direction appears to be toward clearer permissions and more limited connections, such as read-only access. Until banks explain their policies in detail, the best approach is to follow your own bank’s rules and avoid sharing login credentials.

THE BOTTOM LINE

Some U.S. banks are restricting outside AI agents, especially when they sign in using customer credentials or could act without clear approval. But the strongest claims in the latest coverage are not independently confirmed across every bank, and the reports do not establish widespread losses. AI agents may become useful tools for managing money, but for now, keep your banking password private and check with your bank before connecting one.

Sources and Further Reading

• Axios report on AI agents accessing U.S. bank accounts
• Reuters coverage of the Axios report
• Wells Fargo Online Access Agreement update
• Anthropic’s Claude Duet article

Article reflects reporting available on October 23, 2026. Recheck date-sensitive details before publishing later.