Agentic AI

Banks Are Blocking AI Agents: Stop Sharing Your Bank Logins

US banks are blocking AI agents that log in with customer passwords. How GCC and India SMEs should connect AI tools to bank data safely.

Some of the biggest banks in the United States are now blocking AI agents that log in to customer accounts with a username and password. The message for any business using AI in finance is simple: give your AI tools the data they need, through proper channels, and never your bank logins.

This matters well beyond Wall Street. As more finance teams in the GCC and India try AI assistants, the temptation to "just let it log in" will grow. Here is what happened and what to do about it.

What happened

On 22 October, PYMNTS reported that major US banks are finding and removing third-party AI agents working inside customer accounts. Some of these agents appear to have tried actions the customer never asked for.

PaymentExpert, citing an Axios report, named Morgan Stanley, JPMorgan Chase, Bank of America and Citi as restricting access. Banks found agents logging in with customer credentials. Even where the customer had agreed, this broke the banks' terms of service.

Yahoo Finance, republishing International Business Times, reported that some tools contacted bank call centres or did tasks customers never requested. JPMorgan is blocking credential sharing and screen scraping where it can, and Chase has told customers not to share usernames and passwords. Its CEO, Jamie Dimon, put it plainly: "We're not going to allow just any outside agents to come in inside the company."

Why the banks are worried

The core problem is proof. Banks struggle to tell the difference between the real customer, approved software and an agent acting outside its instructions.

Liability is the second problem. If an agent makes a payment nobody asked for, it is not clear whether the loss falls on the customer, the agent provider or the payment firm. JPMorgan's CFO, Jeremy Barnum, warned that customers could face consequences if they handed credentials to an agent that then made an unwanted transaction.

The banks have not shut the door completely. According to PYMNTS, several still allow agents to do read-only work, such as checking balances. Bank of America's CEO, Brian Moynihan, said a human should approve any money leaving an account.

What this means for your business

Your bank in Doha, Dubai, Riyadh or Mumbai may not have made a public statement. That does not make sharing logins safe. Most bank terms already forbid it, and if an outside tool moves money using your credentials, you may struggle to argue the payment was unauthorised.

For a finance team, the lesson splits neatly in two. Reading data is where AI earns its keep: matching bank lines to invoices, spotting unusual payments, building a cash forecast. Moving money is where you want a named person, a second approver and a clear audit trail.

India already has a model for the first part. Under the RBI-regulated Account Aggregator framework, financial data moves between institutions only with the customer's explicit consent. Consent-based, purpose-limited access is the right standard to ask for, wherever you bank.

For example, a Doha trading company that wants an AI assistant to reconcile three bank accounts every morning does not need to give it online banking passwords. It needs a statement feed or an approved bank connection, read-only, for those three accounts.

A checklist for this week

  • List every tool that touches your bank data. Include AI assistants, browser extensions, reconciliation tools and anything a team member signed up for on their own.
  • Ask each one how it connects. An official bank connection or statement feed is fine. "We log in as you" is not.
  • Change any shared passwords. If a login was ever typed into a third-party tool, reset it and check who holds bank tokens.
  • Keep AI on read-only by default. Let agents prepare payment files and suggest matches. Keep release of payments with people, using maker-checker approval.
  • Ask your bank what it supports. Many banks offer statement exports, host-to-host links or approved connections for business clients. Use those first.
  • Read your bank terms. Check what they say about sharing credentials and who carries the loss for unauthorised payments. Check with your adviser if the position is unclear.
  • Review access every quarter. Remove tools nobody uses. Deleting an app does not always end its access, so revoke it at the bank too.

Whichever tool you choose, NexCFO included, ask these same questions before connecting it to your bank data.

AI should read your bank data and help you think; people should still decide when money leaves the account.

This article is general information, not financial, tax, legal or investment advice. Check the details with a qualified adviser before acting on them.

Put an AI finance team on this.

NexCFO AI's specialists handle collections, reporting, forecasting and compliance — and every action waits for your approval.

Book a demo →Meet the AI workforce

Keep reading

WhatsAppBook a demo