Big Banks Are Keeping AI Agents Out of Your Account. Here Is the Safe Way to Use AI on Your Bills.
JPMorgan, Bank of America, Schwab and Citi have all drawn the same line in the past week: an AI can look, but a person moves the money.
By John Martines · October 22, 2026 · 6 min
Four of the biggest US financial firms now stop outside AI agents from logging in or moving money. Here is what that means for a small business that wants AI help with bills, and how to set up the work so a person still presses the button.
The office manager at a landscaping company gets a pitch on a Monday morning. A new AI assistant promises to handle the bills: it reads the supplier invoices in the inbox, logs into the business bank account, and pays them on time, every week, with no one lifting a finger. All it needs is the online banking username and password.
Two weeks ago that might have sounded like a gift. This week, the country's largest banks said, one after another, that they don't want that assistant anywhere near their customers' accounts.
An AI agent is an assistant that does things, not just answers questions. A banking agent is one that signs into an account and acts there: checks balances, moves money, pays bills.
There are two ways such an agent can get in. The first is through a door the bank built and controls, an approved connection that shares only what you allow. The second is by using your username and password and clicking through the bank's website as if it were you. Banks have never liked that second way. Now they are saying so out loud, and blocking it.
How it got here:
- April 29, 2025: Mastercard announced Agent Pay, payment tokens that let an agent pay within limits the cardholder sets.
- October 14, 2025: Visa announced its Trusted Agent Protocol, so stores can recognize approved shopping agents.
- September 30, 2026: Charles Schwab announced Charley, its own AI assistant for clients.
- October 6, 2026: Sierra and Meta published the Personal Agent Protocol, a standard for how a person's agent signs in to a business. We covered it in [Your Next Customer Might Be an AI Assistant](/articles/customer-ai-agents).
- October 13, 2026: on JPMorgan's quarterly results call, chief executive Jamie Dimon said the bank won't let outside agents operate inside its systems. "You don't want rogue agents, and there could be thousands of them," he said. He also warned customers about "a liability transfer" when they hand an agent access.
- October 14, 2026: Bank of America's chief executive, Brian Moynihan, told CNBC the bank strongly believes a human should be in the loop to approve any money moving out of an account. The bank is limiting agents' access to consumer deposits while it works out who is liable for payments an AI approves.
- October 16, 2026: Schwab's chief executive, Rick Wurster, told Bloomberg Television that Schwab blocks outside agents trying to move client money, including Meta's personal agent, Muse. "We can't control it, the client can't control it, something could go wrong and who's liable?"
- October 19, 2026: citing the Financial Times, trade press reported that Citi does not allow third-party AI agents or assistants to log into customer accounts or carry out transactions.
The banks did not ban AI. They drew a line between looking and moving. JPMorgan already lets some outside tools view account data, but not send money or pay bills. Citi still lets approved data-sharing services read account information. Schwab plans secure connections that let a client's chosen AI check balances and do research, but not move money.
Three things follow for an owner.
An AI tool that asks for your banking password is now working against your bank. If it logs in as you and something goes wrong, the bank's position is already on record: you handed over the keys. That is the "liability transfer" Dimon described.
The reporting is about personal accounts. None of the coverage spells out the rules for business accounts, and they may differ. That is a question for your banker, not a guess.
The useful part of AI on bills was never the last click. Gathering invoices, matching them to purchase orders, catching a duplicate, flagging a price that jumped, drafting the week's payment list: that is the hour of work. Pressing "pay" takes a minute, and it is the one step where a mistake or a fraud costs real money.
1. Ask your bank one question this week. Call or message your business banker: "Can an outside AI tool log into our business account, and which apps can read our balances today?" Write down the answer and the date.
2. Never give an AI tool your online banking password. If a bill-pay, bookkeeping or AI assistant asks for it, rather than connecting through a link your bank offers, say no. A tool that signs in as you can do anything you can do, and you carry the loss if it goes wrong.
3. Split the job in two. Let AI do the preparing: read the invoices, match them, draft the payment list with any odd items flagged. A person reviews the list and releases the payments from the bank's own site or app. Our free [Agent Permission Planner](/tools/agent-planner) walks through this one job at a time; moving money always lands in the "a person does it" pile.
4. Turn on the controls your bank already has. Many business banking services offer a second approver for payments over an amount you choose, daily limits, and alerts when a new payee is added or bank details change. Ask which ones you have and switch them on. They protect you from a bad agent and a bad email alike.
5. Clean up the connected apps list. In your bank's online settings, find the list of apps and services that can see your accounts. Remove anything nobody recognizes or uses. Do the same in your bookkeeping software's bank connections.
- How do you connect to our bank: through the bank's approved link, or with our username and password?
- Can you only read our accounts, or can you also move money? Can we limit you to read only?
- When you prepare a payment, who presses the final button?
- If a payment goes to the wrong place because of your tool, who carries the loss: us, you, or the bank?
- What happens to our bill paying if our bank blocks your tool next month?
Banks, card networks and AI companies are now working out the rules for agents that move money: who approves, who is liable, how an agent proves it is legitimate. Bank of America says it is building those protocols, and Schwab and JPMorgan are building their own tools. Until those rules exist, the answer from the biggest banks is clear and it is a good one for any small business to copy. Let AI get the bills ready. Keep the last click with a person.
---
Sources (checked October 20, 2026). Citi, Bank of America and JPMorgan positions, Moynihan's CNBC quote, and Citi's read-only data sharing: "Citi and Bank of America restrict AI agents from accessing accounts" (The Paypers, Oct 19, 2026) and Retail Banker International (Vidhya Edwards Munnangi, Oct 19, 2026), both citing the Financial Times. Dimon's quotes and the October 13 results call: Customer Experience Dive (Oct 14, 2026); call date also reported by PYMNTS and Cybernews. Schwab, Wurster's Bloomberg Television interview, Muse, Charley and JPMorgan's view-only access: "Schwab and JPMorgan are blocking AI agents from moving customer money" (The Next Web, Oct 16, 2026). Mastercard Agent Pay: "Mastercard unveils Agent Pay, pioneering agentic payments technology to power commerce in the age of AI" (Mastercard, Apr 29, 2025). Visa Trusted Agent Protocol: "Visa Introduces Trusted Agent Protocol: An Ecosystem-Led Framework for AI Commerce" (Visa, Oct 14, 2025). Personal Agent Protocol: Sierra, "Introducing Personal Agent Protocol" (Oct 6, 2026).