Banks Now Say AI Agent Losses Are Your Problem. Here Is How We Protect Business Accounts
If you have handed an AI assistant your bank login so it can sort expenses or pay a few bills, the rules just changed under your feet. Several of the largest US banks now say that anything an outside AI agent does in your account counts as your own action. That includes mistakes. It also includes what a hacker does after breaking into the agent.
I run software and automation projects at ConceptRecall, and a growing share of our clients want AI tools touching their money. Store owners want payouts reconciled. Restaurant groups want supplier invoices paid on time. Finance leads want spend tracked without three spreadsheets. But the bank fine print now decides who pays when it goes wrong.
What the Banks Actually Changed
This was not one bank. It is a pattern that built up over about six weeks.
- October 2026. JPMorgan Chase and Bank of America announced policies that put the risk of outside AI agents on the customer. Bank of America asked customers not to share banking passwords with AI assistants. Reporting on the JPMorgan agreement says customers who let agents in “assume the entire risk” of improper credential use.
- Week of November 16, 2026. Citibank updated its Online User Agreement. A Citi spokesperson said the change was meant to “reinforce that customers who choose to grant account access to third-party AI agents are responsible” for that activity. Citizens emailed customers with the same warning.
- November 18, 2026. PNC changed its terms in the same direction, according to TechCrunch.
- November 24, 2026. Wells Fargo’s updated Online Access Agreement takes effect. Per Tech Times, it treats transactions by your delegates, AI agents included, as authorized, even ones you did not want or intend.
Two details stand out. First, the banks say they will not reverse transactions made by an automated assistant, intended or not. Second, Citi makes an exception for its own in-house AI agents. The banks clearly want you using their tools, not someone else’s.
Why This Hits Business Owners Harder
Most coverage this week focuses on personal finance apps. That leaves out the people who should be most concerned, which is anyone running payroll, supplier payments or store payouts through a bank account.
Here is the uncomfortable part. Consumer accounts in the US have Regulation E, which caps what you lose on unauthorized transfers if you report fast. Business accounts usually do not get that protection at all. They run on your commercial account agreement. So the safety net was already thinner for businesses, and an AI agent with your credentials can thin it further. None of the reporting I read says whether these new terms cover commercial accounts, which means you need to check your own agreement rather than assume.
The risk is also bigger in practice. A business account moves more money, more often, to more payees. A personal finance bot that misreads a bill is annoying. An agent that pays the wrong supplier forty times on a Friday night is a real cash flow problem.
A Real-World Scenario
Picture a restaurant owner with three locations. To save time, they connect a general AI assistant to the main operating account with their full online banking login. The assistant reads supplier emails, matches invoices and schedules payments.
Then a fake invoice lands in the inbox, made to look like it came from the meat supplier with new bank details. The assistant does what it was told and pays it. Under the new terms, the bank sees an authorized payment made with valid credentials. The owner is the one chasing the money.
We see versions of this in Shopify and WooCommerce stores too, where an automation tool holds the login for the account that receives payouts. The tool is not malicious. It is just holding a master key it never needed.
The Five-Step Lockdown We Use With Clients
You do not need to ban AI from your finances. You need to give it less power and more supervision. This is the order we work in.
Step 1. List every tool that touches your bank
Write down every app, bot, browser extension and automation workflow that has your banking login or a bank connection. Include bookkeeping software, expense tools, n8n or Zapier flows and any AI assistant you have pasted credentials into.
Step 2. Remove shared passwords first
Any tool holding your actual username and password is the biggest risk. Change that password today and enable two-factor login. If the tool really needs bank data, reconnect it through the bank’s official connection or a regulated data aggregator, which usually gives a revocable token instead of your full login.
Step 3. Default to read-only
Reading transactions is very different from moving money. Most of the value of AI in finance, such as spend reports, categorizing costs and spotting duplicate charges, needs only read access. Some newer AI finance tools are built read-only by design. Give write access only where a human has signed off on the reason.
Step 4. Ring-fence the money an agent can reach
If an agent must make payments, give it a separate account with a set balance, or virtual cards with hard limits per payee and per day. Keep payroll, tax reserves and your main operating balance in accounts the agent cannot see. If something goes wrong, the damage stops at the limit you chose.
Step 5. Keep a human on new payees and big amounts
Set approval rules so any new payee, changed bank details or payment above a set amount needs a person to confirm. Turn on bank alerts for every outgoing transfer. Then review the agent’s activity log once a week. This one step would have stopped the fake invoice in the scenario above.
What I Would Do This Week
Open your bank’s online agreement and search for “agent”, “third party” and “credentials”. If your bank is one of the names above, assume the stricter reading applies to you until your relationship manager says otherwise in writing. Then run steps one and two before the weekend. They take about an hour and close the biggest gap.
Frequently Asked Questions
Do these new bank rules apply to business accounts?
The reporting so far focuses on consumer and personal finance use, and none of it confirms how the terms apply to commercial accounts. Business accounts already have fewer protections than consumer accounts, so check your own agreement and ask your bank directly.
Is it safe to connect bookkeeping or AI finance tools to my bank?
It can be, if the tool uses an official bank connection, has read-only access where possible and never stores your actual login. The danger comes from sharing passwords and giving payment powers without limits or human approval.
Will my bank refund a payment my AI agent made by mistake?
Based on the updated terms from Citi, PNC, Citizens, JPMorgan and Bank of America, you should not count on it. These banks say activity by an agent you gave access to is your responsibility, even when it was unintended or caused by a hack.
The Takeaway
AI can save your finance team real hours, but the banks have now made it clear who carries the risk. Treat every agent like a new employee on day one. Limited access, a spending cap and someone checking their work.
If you would like help auditing which tools touch your accounts, building a safer automation setup, or you need a hand building or fixing a Shopify or WooCommerce store, developing a custom app, or getting unstuck with ads and SEO, our team at ConceptRecall is always up for an honest conversation.