If you let an AI assistant log in to your Citi account and something goes wrong, Citi now says the loss is yours, even if the tool is hacked. The country's third-largest commercial bank updated its online user agreement this week to make customers who grant account access to third-party AI agents responsible for what those agents do, including what criminals do with leaked credentials. TechCrunch reports that PNC changed its terms on November 18 and that Citizens emailed customers warning they are on the hook for anything their AI assistant does. JPMorgan Chase and Bank of America adopted similar positions in October.
There is one carve-out. Citi says customers can use the AI agents the bank itself provides without taking on that extra liability.
What a banking agent actually does
An AI agent is software that carries out tasks for you instead of just answering questions. To track spending, pay bills or move money, many of these assistants ask for your online banking username and password and then sign in as you. From the bank's side, a payment ordered by you, by your agent or by someone who stole your password from the agent looks exactly the same.
That is the core of the banks' argument. In an October email, Bank of America told customers that anyone who shares credentials is responsible for the activity that follows. It listed three risks: the AI may act without your knowledge, the bank may read the activity as an account takeover and lock you out, and a security flaw in the assistant could hand your login to criminals.

Why banks are drawing the line now
Banks see these tools as unreliable and say they will not reverse payments an automated assistant makes, whether it meant to or not. They also see a competitive threat. An agent that shops around could pull money out of low-interest deposit accounts that help fund bank margins, and bank stocks briefly dipped on that prospect.
JPMorgan Chase goes further. According to The Paypers, it blocks outside agents from acting for retail customers and only lets some read data such as balances through aggregators like Plaid, with no transfers or bill payments. Anthropic's personal finance tool, launched in early November, is also read-only and cannot move money.
How to protect your accounts
Do not give your banking username and password to an AI assistant. If you want help with budgeting, pick a read-only tool or one your bank offers directly. For shopping, a setup like ChatGPT Wallet is closer to what banks want, because the agent pays with a tokenized card number, a stand-in that never exposes your real one, and you approve every purchase with Face ID or Touch ID.
If you already linked an agent to your bank, change your banking password and revoke the access you granted. Turn on transaction alerts so you see every charge as it happens, and report anything you don't recognize to your bank right away.
Europe takes a different approach
European rules lean the other way. Under the EU's payment services law, a bank must refund an unauthorized payment immediately and carry the burden of proving the customer was grossly negligent. According to Clifford Chance, the EU Court of Justice ruled on October 15 that even suspected gross negligence does not let a bank delay that refund. The law does not mention AI agents, though, and no rule yet says whether handing your password to one counts as gross negligence.

The conversation starts here
Sign in with a supporter account to comment. Sign in




Nobody has commented yet. Want to go first?