Big banks draw the line at outside AI agents moving customers’ money

JPMorganChase, Citi, Bank of America and Charles Schwab are restricting outside AI agents from moving customers’ money, while leaving some authorised account-reading open. The split is a practical early boundary for AI agents in finance: they may be able to see a balance, but that does not mean they can pay a bill.

Watch Desk analysis

What happened

A PYMNTS report published on 19 October says the four firms set those limits during the same week. JPMorganChase and Citi allow some authorised, read-only access; Schwab says it will build MCP servers for clients’ chosen AI models to check balances and conduct research under Schwab’s rules. Schwab chief executive Rick Wurster said the firm blocks unauthorised agents, including Meta’s Muse, in the tens and twenties each week, according to Bloomberg as cited by PYMNTS.

Bank of America is working with agent developers on protocols and liability rules. Chief executive Brian Moynihan said a person should remain in the loop before money leaves an account, the report says. The article also describes banks developing their own assistants and money-moving features, including JPMorgan’s Smart Cash and Schwab’s Charley.

Why it matters

The distinction is between letting an agent retrieve financial information and giving it authority to act on that information. For customers, the immediate takeaway is that access to balances does not necessarily mean an outside assistant can transfer funds or pay bills. For developers, permissions and responsibility for a transaction are not details to leave until after the demo.

The banks’ stance is not a blanket rejection of AI in finance. They are keeping the door open to controlled, read-only connections while reserving money movement for their own systems or for arrangements with clearer rules. PYMNTS also cites survey findings that 37% of AI power users treated AI as their main tool for managing finances and banking, and that consumers want permission before an agent completes a purchase. Those figures provide context, not proof that every customer or bank has the same view.

Our read

This is a sensible boundary, even if it is also a useful way for banks to keep customers inside their own apps. Reading an account and moving its money are different powers; bundling them together because an agent can do both would be a rather bold shortcut. Customers should check what an AI connection can actually access and whether it can take action before linking it to an account.

What to watch

• Whether banks publish clear rules for authorised agents and the actions they may take.
• How liability is assigned when an AI agent initiates a payment.
• Whether bank-built assistants gain money-moving features before outside agents do.

Discussion spark: Should banks keep money movement under their own control, or should customers be able to authorise outside AI agents to pay and transfer on their behalf?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.