Short answer: Read-only access through a bank-approved connection is reasonable for most small businesses. Handing an AI agent your actual login, or letting it move money without a human approving each payment, is risky right now, and the biggest US banks are blocking it.
What's happening: banks are finding AI agents inside customer accounts
On October 22, 2026, Axios reported that major US banks, including Morgan Stanley, Bank of America, JPMorgan, and Wells Fargo, are detecting unauthorized third-party AI agents operating inside customer accounts. Some of those agents appeared to take actions independent of what the user asked for. In most cases, the customer had handed the agent their own login, which goes against most banks' rules that humans log in to their own accounts. Bank executives told Axios their biggest fear is what an agent might do if a user has given it purchase authority or the ability to move funds. Some banks also struggle to tell whether account activity came from an authorized agent or one that has gone rogue. (Reuters, which summarized the Axios story, said it could not immediately verify the report.)
The banks have been drawing a firm line all month:
- Charles Schwab CEO Rick Wurster said the firm blocks unauthorized AI agents a double-digit number of times every week, as PYMNTS reported.
- JPMorgan Chase CEO Jamie Dimon said on the bank's October 13 earnings call that it won't let outside agents inside because "you don't want rogue agents, and it could be thousands of them," according to CNBC. Agents can still see balances through approved connectors like Plaid, but they can't send money or pay bills. Dimon added that people using these agents "should understand there's a liability transfer."
- Bank of America CEO Brian Moynihan said there should be "a human in the loop to authorise money moving out of your account," per Retail Banker International.
- Citi said it does not permit third-party AI agents to log in or transact, and allows limited read-only access through approved open banking providers.
Regulators are paying attention. On October 13, Reuters reported that US regulators, including the Federal Reserve, OCC, and FDIC, are asking banks what limits they place on AI agents, whether humans stay involved in key decisions, and how they can shut those systems down.
Attackers are using AI agents too. On October 8, Reuters reported that CrowdStrike believes a China-based suspect likely used AI agents in recent cyberattacks on South Korean banks. At least 9 South Korean banks have disclosed or been reported as targets since late September, and one, Shinhan Bank, said personal information of about 25,000 customers was compromised.
What this could cost your business
If the largest banks in the country are asking "who approved this payment?", your business should be asking it too. Here is how the risk shows up for an owner:
- Payments you never approved. An agent with your login and bill-pay access can pay the wrong invoice, pay one twice, or send money to a new payee. Axios reported that some agents have already tried actions their users never requested.
- The liability may land on you. Most bank agreements say you are the only one who logs in. If you shared your password and money goes out, the bank may point to its terms. Dimon's "liability transfer" comment is the plain version of that warning.
- Weeks to get money back. Disputes, fraud claims, and vendor refunds take time. Every week that money is stuck is a week it isn't available for your business.
- Cash flow and payroll pressure. A single bad payment of a few thousand dollars can turn a comfortable week into a tight one, especially right before payroll or a big card payment.
- Losing access without warning. If your bookkeeping or bill-pay workflow depends on an agent logging in as you, it can break the day your bank tightens its rules. Several large banks just did.
- Data exposure and compliance headaches. Your bank data includes customer payments, employee pay, and vendor details. A tool that stores your login or your data poorly becomes a target, and a breach can bring notification duties, insurer questions, and lost trust.
Why clean, protected financial data is the foundation of financial health
Every decision you make about your business runs on your numbers. When those numbers are accurate, protected, and only touched by people and tools you trust, you can act with confidence. When anyone or anything can change them, every report built on top of them is in question.
Here is what good data integrity gives you:
- Better decisions. Hiring, pricing, and how much you can pay yourself all depend on knowing what really came in and went out.
- Reliable cash forecasts. A forecast is only as good as the transactions feeding it. One unexplained payment can throw off the next 13 weeks.
- Trust from lenders and buyers. Banks and future buyers want books they can follow from start to finish. Gaps, mystery transactions, or payments nobody can explain slow down financing and lower what a buyer is willing to pay.
- Clean books every month. Controlled access means fewer surprises at month-end and less time spent cleaning up.
- A clear audit trail. When every payment shows who prepared it and who approved it, you can answer any question from your CPA, a lender, or the IRS quickly.
This is the part of AI adoption most owners skip, and it's where a trusted advisor earns their keep. At Black Lotus Financial Group, we help owners decide which tools can see their numbers, which can touch their money, and how every dollar gets approved, so the data behind their decisions stays something they can count on.
7 safeguards that let you use AI and keep control of your cash
You can get most of the time savings from AI while keeping a human hand on every dollar that leaves. Here is the setup we recommend to owners:
- Never give an AI tool your bank username and password. Use only bank-approved connections (the "connect your bank" flow through providers like Plaid, or your bank's own integrations). If a tool asks you to type your login into its own screen, treat that as a red flag.
- Default to read-only. Let AI read balances and transactions so it can categorize, reconcile, and forecast. That covers most of the value for a small business.
- Keep a human approval on every payment. Let the AI prepare the bill-pay list or draft the payment. A person approves it inside the bank's own app.
- Turn on dual approval for larger payments. Set a dollar threshold that fits your business, and require a second person to approve any payment above it. Most business banking platforms have this setting built in.
- Set up account alerts. Text or email alerts for any outgoing payment, any new payee, and any login from a new device. An alert is often the fastest way to catch a mistake before it grows.
- Review and revoke access every quarter. Check your bank's connected-apps list and your accounting software's integrations. Remove anything you don't use or don't recognize. Change your password if you ever shared it.
- Vet the vendor before you connect. Ask how it connects to your bank, whether it can move money, who is liable for errors, where your data is stored, and how to shut it off. If the answers are vague, wait.
Your CFO should own this list. A good one treats AI tools like a new hire with access to the checkbook: clear limits, a second set of eyes, and a regular check on what they've been doing.
FAQ
Can AI agents legally access my bank account?
If you give an agent your login, it can usually get in, but most banks' terms require you to log in yourself. Banks including JPMorgan Chase and Citi now block third-party agents from logging in for customers and allow read-only access through approved providers.
Is read-only access to my business bank account safe?
Read-only access through a bank-approved connection is the lower-risk option, because the tool can see your data without moving money. You should still vet the vendor and review connected apps regularly, since your financial data is sensitive.
Who is liable if an AI agent sends a payment I didn't approve?
It's unsettled. Banks and regulators are still working out liability, and JPMorgan's CEO warned users to understand there's a "liability transfer" when they authorize outside agents. Assume the risk may land on you if you shared your credentials.
Should a small business use AI for bookkeeping and bill pay?
Yes, with guardrails. AI is useful for categorizing transactions, catching duplicate bills, and forecasting cash. Keep payments behind human approval, and set a dollar threshold above which a second person must approve.
What should I do if I already gave an AI tool my bank password?
Change your bank password today, turn on multi-factor authentication, review recent transactions and payees, and reconnect the tool through a read-only, bank-approved connection if you still want it.