Five US Banks, Including Citi, Shift the Risk of Hacked AI Agents onto Customers
Five US banks — Citi, JPMorgan, Bank of America, PNC and Citizens — have warned customers that if they give a third-party AI agent their login and password, the customer is responsible for what the agent does. Citi went further: if the agent leaks your data or gets hacked, the bank will not return the stolen funds. Picture a budgeting assistant that mistakenly pays the wrong bill: the bank is not obligated to reverse the payment. That is why fintech startups are better off starting with a read-only mode.
AI-processed from TechCrunch; edited by Hamidun News
Imagine giving an app with an AI assistant your login and password for your online banking account. It analyzes your spending, advises where to invest, and pays your bills on its own. It's convenient, until there's an error or a hack. This week it became clear who pays in such a case: five major US banks, namely Citi, JPMorgan, Bank of America, PNC and Citizens, have told customers that the responsibility lies with them, not the bank.
What Citi changed
Citibank, the third-largest commercial bank in the US, updated its user agreement. If a customer has given a third-party AI agent access to their accounts, they are responsible for everything the agent does. The bank is not obligated to reverse such a payment, regardless of whether it was intentional or the result of an error by an automated assistant.
There is also a tougher clause. If an AI service leaks a customer's credentials or the service itself is hacked, the customer is also responsible for the actions of the attackers with the account. A Citi representative told TechCrunch that the bank wants people to understand the risks of handing logins and permissions to such tools.
Citi is not alone. JPMorgan and Bank of America, the country's two largest banks, announced similar rules last month. PNC, the eighth-largest bank, changed its terms today. Citizens sent customers a letter stating that they are responsible for the actions of their AI assistant.
Why banks are doing this
The first reason is security. An agent with an account password becomes a new point of failure. It can make a mistake in a payment, it can be deceived, and the service that holds the data can have its access credentials stolen. Previously, the bank handled such disputes itself; now it is shifting the risk onto the customer in advance.
The second reason is competition. According to TechCrunch, markets briefly grew nervous when investors realized that AI tools could help people avoid unnecessary fees and move money out of low-interest deposits. Part of bank margins rests on exactly that. If agents become a routine way to make financial decisions, banks will want to control this channel themselves. The caveat in Citi's terms is telling: agents offered by the bank itself can be used without additional liability.
What this means for developers and product managers
For fintech startups and AI labs looking to enter personal finance, this is an unwelcome signal. The niche is lucrative: fees circulate there, and there is room to deliver real value to customers. But people already reluctantly give agents access to their email and correspondence. A bank account is the next level of distrust, and now the customer also has no protection from the bank.
Anthropic apparently took this into account. Its personal finance tool, launched this month, is described as "read-only": it cannot transfer money, manage it, or invest. For those building similar products, there are several practical takeaways:
- start with read-only and analytics functions, and add account operations later, with separate human confirmation;
- do not ask users for their password if there is a narrower way to access data with limited permissions;
- explain clearly in the interface who is responsible for errors and leaks;
- monitor your contracts with banks: they determine whether your agent can work with accounts at all.
TechCrunch does not say whether banks will allow third-party agents through official interfaces, or on what terms. But the direction is clear: AI labs promise that agents will change how people interact with services, while the services answer "wait a minute." It was the same with taxis and food delivery. How the dispute with banks ends will largely determine which investments in AI agents pay off.
- Nov 21, 2026A Developer Had Claude Install 900 Apps and Built a Free Mac Cleaner
- Nov 21, 2026Apple Is Building an AI Agent and Making It Harder for Outsiders to Get Access: Permission Now Takes 14 Steps, Not 4
- Nov 21, 2026Qwen3-0.6B Fine-Tuned on a Mac mini: Errors Dropped to 3 out of 30 from 9
- Nov 21, 2026The Programmer Stopped Being an AI Agent Dispatcher: Multica Takes Over the Handoff Routine
Key Questions Answered
What Citi changed?
Citibank, the third-largest commercial bank in the US, updated its user agreement. If a customer has given a third-party AI agent access to their accounts, they are responsible for everything the agent does. The bank is not obligated to reverse such a payment, regardless of whether it was intentional or the result of an error by an automated assistant. There is also a tougher clause.
Why banks are doing this?
The first reason is security. An agent with an account password becomes a new point of failure. It can make a mistake in a payment, it can be deceived, and the service that holds the data can have its access credentials stolen. Previously, the bank handled such disputes itself; now it is shifting the risk onto the customer in advance. The second reason is competition.
What this means for developers and product managers?
For fintech startups and AI labs looking to enter personal finance, this is an unwelcome signal. The niche is lucrative: fees circulate there, and there is room to deliver real value to customers. But people already reluctantly give agents access to their email and correspondence. A bank account is the next level of distrust, and now the customer also has no protection from the bank. Anthropic apparently took this into account.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.