AI·News & analysis
Citi and PNC join the banks telling customers: your AI agent, your risk
Citi, PNC and Citizens have told customers that if an AI tool they connect to their account errs or gets hacked, the loss is theirs. JPMorgan and Bank of America did the same last month.

Tide
Wave
Sci-fi
4/10
Reality
Shipping
Your robot assistant needs a signed permission slip.How we rate
Citi told customers that anything a third-party AI agent does with their account is their responsibility, including damage after a hack or a leak of login details.
PNC and Citizens followed this week, after JPMorgan and Bank of America took the same line in October. The banks fear fraud and the loss of cheap deposits. Regulators privately welcome the limits, and officials say the liability rules may need legal clarification.
What to know
- Citi changed its Online User Agreement this week: customers who give a third-party AI agent access to their account are responsible for what it does.
- Citi's update also says that if the AI tool leaks a customer's login details or is hacked, the customer answers for what hackers do with the account.
- JPMorgan and Bank of America announced similar policies last month, PNC changed its terms on November 18, and Citizens emailed customers this week.
- Banks still allow read-only access to balances, and some are building their own assistants for the tasks they will not hand to outsiders.
Citi told customers this week that if they hand their bank login to an AI tool, whatever goes wrong is on them. That includes the case where the tool itself gets hacked.
Citi is not alone. Here is the part that sounds like a group chat: one by one, the biggest US banks are sending the same message.
What exactly did Citi change?
Citi is the third largest commercial bank in the United States. It updated its Online User Agreement, the rulebook customers accept when they use its online banking.
The new wording says customers who let third-party AI agents into their accounts are responsible for what those agents do. An AI agent is software that acts for you instead of only answering questions. It can read your spending, pay a bill or move money, if you give it your login details.
A Citi spokesperson explained the thinking. "As third-party AI tools become more widely available, we want customers to understand the risks associated with sharing banking credentials or authorizing those tools to access their accounts," the spokesperson said.
Credentials just means your username and password. The update goes one step further than most. If the AI tool leaks them or is hacked, you are responsible for anything the hackers then do with your account.
Which other banks have done this?
Citi is the latest in a short line. The pattern, in order:
- October 13: JPMorgan CEO Jamie Dimon said on the bank's earnings call that outside agents would not get inside the bank. He warned about "rogue agents" and said customers should understand "the liability transfer."
- October 14: Bank of America said it is limiting agent access to consumer deposits while it writes rules about who pays when an AI-approved payment goes wrong. CEO Brian Moynihan said a human should stay in the loop to approve money leaving an account.
- Mid-October: Charles Schwab said it blocks big outside agents, including Meta's Muse, "in the tens and 20s" each week, according to CEO Rick Wurster.
- November 18: PNC, the eighth largest US bank, changed its terms.
- This week: Citizens, another large bank, emailed customers that they are on the hook for anything their AI assistant does.
By the numbers: five big names in about five weeks, all with the same core message.
What does "liability" mean here?
Liability is a plain idea: who pays when something goes wrong. Normally, if a thief drains your account, banks have rules for refunds.
An AI agent blurs that. You gave it permission, so the bank can say the payment was authorized by you. The agent may make a wrong payment on its own, and the bank says it will not reverse it.
JPMorgan's terms of use go the furthest. They say the user is responsible for what an agent does, even if it performs unintended actions or goes beyond the authority you gave it.
In real life You tell a budgeting app to pay your electric bill. A bug sends the payment twice. Under these terms the bank treats both payments as yours, and getting the money back is your problem.
Why are the banks doing this?
Security is the first reason. A login given to an outside tool sits on someone else's servers, and a tool that holds many logins is a juicy target. JPMorgan has said it worries about rogue agents that could number in the thousands.
Money is the second reason, and it is bigger than it looks. An economist at the investment firm Apollo warned that AI assistants could automatically sweep household cash into accounts paying between 3.3% and 5%. The average US checking account pays about 0.1%.
The numbers are large:
- Deposits: JPMorgan, Bank of America and Wells Fargo together hold $1.6 trillion in non-interest-bearing deposits, about 16% of their liabilities.
- The cost: paying just 3% on those balances would cost the three banks about $47 billion a year.
- The comparison: that is close to half of their combined annual earnings.
Banks use cheap deposits to fund their loans. If agents move that cash away, lending gets more expensive.
The catch: not every bank leader sees it as a crisis. Wells Fargo CEO Charles Scharf said high-rate accounts have existed for a long time, and anyone can move money to one online in no time. Moynihan said people mostly keep checking balances to pay bills, not to earn interest.
Where do the AI companies fit in?
AI companies are racing into personal finance. A finance tool lets them add real value for customers, and it opens a new stream of fees. Meta launched its Muse personal assistant in September, and that added to the worry about bots shifting large amounts of cash.
Anthropic chose a careful design for the personal finance tool it launched earlier this month. It is billed as "read-only" and it "can't move, manage, or invest your money."
That is the gap banks are happy to live with. Read-only access means a tool can look at your balances but cannot send a cent.
Banks also see a competitive threat. If agents become the main way people handle money, banks want to control that path. Markets briefly shuddered at the idea that AI tools could help people dodge fees or leave low-interest accounts.
What are the banks building instead?
Banks are not saying "no AI." They are saying "our AI."
- JPMorgan: it is building tools of its own, and Dimon's 2025 shareholder letter described a feature called Smart Cash. It would move client money between checking accounts and higher-paying investment products.
- Schwab: it announced an assistant called Charley on September 30. It answers questions on transfers and trading, and hands off to a person when needed.
- Citi: its terms say customers can use AI agents provided by Citi without worrying about extra liability.
- Bank of America: it is working with agent developers on rules for how an agent may touch an account, so customers never find money gone without knowing where it went.
Schwab also plans to let a customer's chosen AI model check balances and run research, under rules Schwab sets. Moving money stays off the table. "We're not going to allow them to move money," Wurster said.
What do regulators say?
Not much on the record. Wall Street executives have contacted the Federal Reserve about the fast rise of consumers using AI agents for money tasks, according to a person familiar with the discussions. The Fed declined to comment.
Regulators have privately welcomed the bank restrictions. Officials also believe the law may need to say how customers authorize an autonomous system to act for them, and who is liable when something goes wrong. For now the banks are filling that gap with their own contracts.
What does this mean for AI beyond banking?
Banks are one example of a wider pattern. AI labs promise that agents will change how people deal with all kinds of companies. Then the companies reply with "hold on a minute." It has already happened with car-hailing and food service.
How these fights end will help decide which AI investments pay off. An agent that cannot act on your behalf is just a very polite chatbot.
For more on the agent boom behind this, see our story on Meta's Muse for small businesses.
What it means for you
- Check your bank's terms: if you use an AI tool that logs in for you, read your bank's updated user agreement first.
- Prefer read-only: a tool that can only look at balances cannot make a costly mistake.
- Never paste your bank password into a chatbot: under Citi's terms, a leak or hack is on you.
- Ask your bank: many are launching their own assistants, and those come without the extra liability warning.
The bottom line
Banks are drawing a clear line: you can let an AI look, but if you let it act, you carry the risk. The reasons are security, plus a fight over who controls your money.
Expect the rules to be rewritten as regulators weigh in and banks ship their own assistants. Until then, treat an AI agent with your bank login like a new house guest with a spare key: handy, but you are responsible for the house.
Key facts
- Latest bank
- Citi, with PNC and Citizens the same week
- Earlier banks
- JPMorgan and Bank of America, October 2026
- What changed
- Customers answer for what their AI agent does
- Still allowed
- Read-only access to balances through approved connectors
- Deposits at stake
- $1.6 trillion held without interest at three big banks
Got questions?
Quick answers, plain wordsWhat did Citi change?
Citi updated its Online User Agreement. It now says customers who give third-party AI agents access to their accounts are responsible for activity done through those agents, including after a leak or hack of the AI tool.
Which other banks have said the same?
JPMorgan and Bank of America announced similar policies in October. PNC changed its terms on November 18, and Citizens emailed customers this week with the same warning.
What is an AI agent?
An AI agent is software that takes actions for you, not just answers questions. In banking, it could read your spending, pay a bill or move money if you give it your login details.
Can an AI agent still see my balance?
Often yes. Banks such as Citi and JPMorgan let approved connectors share account information in read-only form, which means the tool can look but cannot move money.
What happens if my AI tool is hacked?
Under Citi's updated terms, you are responsible for what the hackers do with your account. Citi says that is the risk of sharing banking credentials with outside AI tools.
Does this apply to the bank's own assistant?
No. Citi's terms say customers can use AI agents the bank provides without worrying about extra liability. JPMorgan and Schwab are building their own assistants too.
Why do banks care so much?
There are two reasons. One is security: a stolen login or a wrong payment is hard to undo. The other is money: agents could move idle cash into accounts that pay more, which would cost banks cheap funding.
Are there safer ways to use an AI tool with my money?
Read-only tools are the safest. Anthropic's personal finance tool, launched earlier this month, is billed as read-only and cannot move, manage or invest your money.
Are regulators involved?
Wall Street executives have contacted the Federal Reserve, which declined to comment. Regulators have privately welcomed the bank restrictions, and officials think the law may need to say who is liable when an agent errs.
SourcesTechCrunch
Topics and tagsAI agents, Cybersecurity, ai agents, banking
Go deeper
Related stories

World launches a World ID suite for banks to fight deepfake fraud
Unveiled at Money20/20, World's first industry-specific product bundles its ID app, a Zoom human check and AgentKit so banks can confirm a real person is behind a transfer, a call or an AI agent.

Microsoft Execution Containers now give AI agents a locked room
Microsoft says MXC, its policy-driven way to fence in AI agents on Windows, macOS and Linux, is now generally available, with GitHub Copilot, OpenAI Codex and Replit already on board.

South Korea's president says AI was likely used in a wave of bank hacks
Shinhan, KB Kookmin, Hana and four other lenders reported breaches in a week. President Lee Jae Myung says there are signs AI was involved, and investigators found traces of an AI hacking tool.
More in brief
- ASELSAN's EJDERHA electromagnetic weapon disables 51 drones with one commandNov 21
- Microsoft To Do will shut down on June 30, 2027Nov 21
- Medtronic's Hugo surgical robot cleared for general surgery in the USNov 21
- Google's Gemini now turns noisy fire radio into text for Watch DutyNov 21
- Internet Archive staff move to unionize with CWANov 20
- Etsy sellers are leaving over new terms that allow AI trainingNov 20