Citibank, the third-largest commercial bank in the United States, updated its online user agreement this week. Under the change, customers who give third-party artificial intelligence tools access to their bank accounts are responsible for activities conducted through those tools. That includes cases in which a tool leaks credentials or is hacked and attackers use the resulting access.
AI agents used for personal finance can be given credentials for bank accounts and other financial services. They can then analyze spending and investment decisions, offer advice, and even carry out transactions and pay bills. Banks, however, are concerned about the reliability of automated assistants and are telling customers they will not be responsible for reversing transactions the agents make intentionally or accidentally.
A Citi spokesperson told TechCrunch that the updated agreement was intended to clarify the risks of sharing banking credentials and authorizing outside AI tools to access accounts. The change also reinforces that customers who grant such access are responsible for activity conducted through the agents.
Other banks adopt similar rules
JPMorgan and Bank of America, the two largest banks in the United States, announced similar policies last month. PNC, the eighth-largest bank, changed its terms today. Citizens also emailed customers this week to warn that they are responsible for what their AI assistants do, according to an email seen by TechCrunch.
Citibank's rule is particularly broad: it places responsibility on customers not only for an agent's actions, but also for actions taken by attackers using credentials that have leaked or been obtained in a hack.
Trust and competition in financial services
Personal finance is an attractive field for AI startups seeking new sources of revenue. Such services can help customers manage their finances while giving developers an opportunity to share in the sector's fee income. Access to bank and brokerage accounts, however, requires a particularly high level of trust, especially among consumers already reluctant to share emails or chat records with an AI agent.
Anthropic's personal finance tool, launched earlier this month, takes a different approach: it is explicitly read-only and, according to the company, cannot move, manage or invest a user's money.
Beyond cybersecurity risks, banks also see AI companies entering financial services as competitors. Markets briefly reacted nervously to the prospect that AI tools could make it easier for consumers to avoid unnecessary fees or low-interest deposit accounts that contribute to bank margins. If AI agents become an important part of financial decision-making, banks are likely to fight for control of that point of access. Citibank's terms say customers can use agents supplied by the bank without concern about additional liability.
The banks' response fits a broader business pattern: while AI developers say their agents can transform how customers interact with service providers, the providers themselves often urge caution. Similar tensions have emerged in car-hailing and food service, and the way these relationships develop may help determine which AI investments pay off.



