AI assistants want into your bank account. The banks just rewrote their terms so that whatever the agent does is on you. Two-factor authentication won't save you, and the security work that would is the job of the people writing those terms.
November 14, 2026
Online banking user agreement, updated November 2026
H. Use of Third-party Artificial Intelligence ("AI") Agents. These terms apply to your use of a third-party AI agent or agentic artificial intelligence to access your accounts or use digital services through Citi Online or Citi Mobile.
You are cautioned against using any third-party AI Agent to access your accounts or use any digital services including, but not limited to, use of such agent to provide any information or instruction to Citi or perform any transaction…
If you disregard this caution and provide your Online Credentials to any third-party AI Agent or otherwise permit such agent access to your accounts…
“Citi will consider any access and actions taken by the agent as authorized by You.”
Citi shall have no responsibility or other obligation to investigate the authority or validity of any such actions even if the actions taken by the agent… conflict with any prompts or other instructions provided by You…
Agreed.
Whatever an AI agent does with your login now counts as something you did. Most people click this button without reading the newest section.
The terms changed while you were busy
If you bank online with Citi, the agreement you clicked through has a new section. Section H, added in November 2026, is titled Use of Third-Party Artificial Intelligence ("AI") Agents. It doesn't ban AI agents. It "cautions" you against them, then spells out what happens if you use one anyway: Citi will treat anything the agent does as done by you.
Citi isn't alone. JPMorgan, Bank of America, Schwab and Morgan Stanley have all moved to block or limit outside agents. JPMorgan's Jamie Dimon told analysts that customers who let an outside agent in need to understand the "liability transfer" they're agreeing to. Bank of America emailed customers warning them they'd be responsible for what an agent does in their account.
The pattern is the same at every big bank. The AI companies are asking to get closer to your money. The banks are answering with paperwork that puts the risk on you.
AI companiesBanks
ChatGPT lets Pro subscribers link bank accounts through Plaid, read-only.
Meta's Muse agent adds Plaid account connections.
JPMorgan says on its earnings call it will limit outside agents' access.
Citi and Bank of America announce limits. Schwab says it blocks agents that log in with client credentials.
Morgan Stanley is reported to be blocking outside agents too.
Citi adds Section H to its online banking agreement.
Wells Fargo's new terms take effect: you must be able to prove you revoked an agent.
Two different doors
When an AI "connects to your bank," it can mean two very different things. The pitch for the second door is convenience: let the agent handle it. The fine print says that when the agent handles it badly, you handle the loss.
Door one: a window
A read-only data connection. You sign in on Plaid's screen, not the AI's, and the AI gets a token that lets it see balances and transactions but not move money. You can revoke it.
ChatGPT added this in May through Plaid, which reaches more than 12,000 financial institutions. Meta's Muse followed in September. Citi reportedly still allows limited read-only access through certain open-banking providers.
Looks, can't touch
Door two: your keys
Giving an AI assistant or browser agent your actual username and password, or letting it run inside a session you've already logged into, so it can click around the bank's site the way you would: pay bills, move money, change settings.
This is the door the banks are shutting, and it's the one Citi's Section H is about.
Everything it does is yours
What you're opening yourself up to
Read Citi's Section H closely and you can list the exposures it makes yours.
Tap or press a card to turn it over.
One honest caveat: I couldn't find a documented case of a customer losing money to an AI agent yet. Schwab's CEO says the firm blocks big outside agents in the "tens and twenties" each week, and also that few clients seem interested in handing over their credentials. These clauses are getting written before the losses, which is exactly the right time to read them.
Two-factor is a toll booth, not a guard
Everyone who banks online knows the routine: password, then a six-digit code by text, then the code expires because the text came late, then do it again. It's sold to us as the thing that keeps our money safe.
It isn't pointless. It stops a lot of crude password-guessing. But the version most of us get, a code by text message, is the weakest kind there is. Criminals get around it with SIM swaps, talking your phone carrier into moving your number to their phone. App approvals are better, but they have their own trick, MFA fatigue: prompt after prompt until a tired person taps "approve."
Here's the bigger problem, and it's the one that matters for AI. Two-factor checks you once, at the front door. After you're in, it steps aside. An agent holding your password, or riding along in your logged-in session, is already past the door. Customers get all the friction of two-factor and none of the protection against the new risk.
1,611SIM-swap complaints to the FBI in 2021
$68M+in reported losses that year
Where the burden really belongs
The contracts move the liability to you. They can't move the ability to fix it, because only the institution controls the systems. You can't build a safer login screen for your bank. You can't make it tell a human apart from an agent. You can't give an assistant permission to pay one bill without also giving it the keys to everything else. Those are engineering decisions for the people who run the systems being connected: the banks, the card networks, the data connectors and the AI companies.
Right now those systems aren't built for agents. People who work with the banks say they can't reliably tell whether they're dealing with a customer or that customer's AI. When one survey asked fraud and risk leaders who's responsible when an agent-driven transaction goes wrong, the answers were all over the map.
AI provider
39%
Customer
20%
Merchant
14%
Bank
11%
Darwinium survey of fraud and risk leaders, as reported by Digital Transactions. Remaining responses weren't reported.
Only one in five picked the customer. The bank agreements picked the customer anyway.
The referee isn't on the field yet, either. The federal open-banking rule (Section 1033) that was supposed to set ground rules for sharing your data with outside apps was blocked by a federal court and is being rewritten. A revised proposal went to the White House for review in August and, as of October, was still there.
The banks have a fair argument. An outside agent they can't see, vet or stop is a real security risk, and PNC's CEO called agents doing customers' banking "incredibly unsafe." But there's a business motive too: banks earn money on balances customers leave sitting at low interest, and an agent that moves idle cash to wherever it earns more is a threat to that.
A bank can be right that a stranger with your password is dangerous, and still be the only party that can build a safer way in: limited, revocable permissions for an agent, logins that can't be phished, and clear records of who did what. Until the banks build that, the liability clause is the only thing they've built.
What to do now
Never give an AI your bank password
Not to an assistant, not to a browser extension, not "just this once." That's the act the new terms turn into your liability.
Use the read-only door for AI help with money
Choose a linked connection where you sign in on the bank's or connector's own screen and the AI can look but not move money. Check what you've connected, and disconnect what you don't use.
Do the moving yourself
Let the AI help you decide, then log in and make the payment or transfer with your own hands.
Get off text-message codes where you can
If your bank offers an authenticator app, a passkey or biometric login in its own app, use that. Ask your phone carrier for a PIN or port-out lock on your number.
Already gave an agent your login? Revoke it in writing
Change your password, review recent activity, and tell your bank in writing (secure message, not a phone call) that the agent is no longer authorized. Keep a dated copy.
Read the agreement you actually have
Each bank draws its lines differently between viewing, logging in and paying. The update notice is in your inbox or on your login screen, and the part that matters is usually the newest section.
The bottom line
The AI companies are building doors into your accounts. The banks are building contracts. The only party the new terms make responsible is the customer, who has the least control over how any of it is built. Read the fine print, keep your keys, and keep asking the people who run these systems when they plan to build a safer way in.
This article is general information, not legal or financial advice. For questions about your own accounts, contact your bank or an attorney.
Sources
Citi Online & Citi Mobile User Agreement, Section H: Use of Third-Party AI Agents (November 2026)