In the space of one week, four of the biggest names in American banking drew the same line for AI agents. JPMorganChase, Citigroup, Bank of America and Charles Schwab each said outside agents cannot move customer money, PYMNTS reported on Monday, October 19. A customer who hands a login to an agent now gets turned away at all four, even as those same customers increasingly treat AI as their main tool for managing money. The banks are not rejecting agents outright. They are rejecting agents they cannot control, with credentials they did not issue.
The statements came in quick succession. JPMorganChase chief executive Jamie Dimon warned of rogue agents on the bank's third-quarter earnings call on Tuesday, October 13, and said outside agents would not get inside the bank, adding that customers who use them anyway should understand there is a liability transfer. Schwab chief executive Rick Wurster said the brokerage blocks the big unauthorized agents, including Meta's Muse, in the tens and twenties each week, Bloomberg reported on Friday, October 16. Bank of America chief executive Brian Moynihan said on Wednesday, October 14, that a human should stay in the loop to approve money moving out of an account. By Friday, Citigroup had drawn the same line: third-party agents may not log into accounts or transact.
What stays open is the read-only door. At JPMorganChase, an outside tool connected through an authorized connector such as Plaid can still see balances, but it cannot send money or pay bills, and the bank gave no timeline for changing that. Citi lets certain authorized open-banking providers retrieve account information in read-only form. Schwab plans model context protocol servers so a client's chosen AI model can check balances and run research under rules Schwab sets, while still not moving money. Agents can look. They cannot act. That distinction is now the industry's default architecture, whether agent builders planned for it or not.
The money-moving job itself is not going away; the banks plan to keep it for their own assistants. JPMorganChase is testing Smart Cash, which would move client money automatically between checking accounts and higher-yielding brokerage products inside its own app, the same sweep an outside agent would offer. Schwab is rolling out Charley, the generative AI assistant it announced on September 30, which answers questions on transfers, trading and account actions and hands off to a Schwab representative. The demand is real: PYMNTS Intelligence found in February that 37 percent of AI power users relied on a native AI platform as their primary tool for managing finances and banking in December, and a separate June report found consumers want an agent to ask permission before completing a purchase.
For anyone building agents, that permission step is the design spec, not a courtesy. Bank of America says it is working with agent developers on protocols that decide which party is liable for a payment an AI authorizes, so customers do not find money gone without understanding why. One firm has already shipped the pattern the banks describe: Robinhood lets outside agents trade through its banking MCP server with a customer-set spending limit, optional manual approvals and no access to the primary card number. Build with scoped, revocable credentials, explicit approval for irreversible actions, and a clear answer to who pays when the agent is wrong. An agent that asks for a customer's password is not being bold. After this week, it is simply locked out.
