Image: Citi
Citi has told customers that if they hand their banking login to a third-party AI agent, they carry the loss. The bank updated its Online User Agreement this week to add a section on outside AI agents, and the terms cover more than bad advice: if the tool leaks the credentials and someone else uses them, Citi treats that activity as authorized by the customer too.
A Citi spokesperson described the change as clarity rather than a new restriction. "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. The bank says customers who choose to grant access are responsible for what happens through those agents.
Citi is not first. JPMorgan and Bank of America, the two largest US banks, brought in similar policies last month, and PNC, the eighth largest, changed its terms on the same day as Citi. Citizens emailed customers this week with a comparable warning, according to TechCrunch, which saw the message. The pattern is consistent: the bank treats the agent's actions as the customer's own instructions.
The carve-out matters as much as the rule. Citi's terms apply to outside software, not to AI agents the bank provides inside its own app, which customers can use without extra liability. That keeps the bank's own assistant on one side of the line and every startup tool on the other.
There is a reason banks are nervous. An agent that holds your login can read balances, move money and pay bills, and the bank cannot tell a careful instruction from a confused one. Citi's agreement also says it does not have to investigate whether an action was within the authority the customer gave the agent, and that it can cut off an agent's access at any time without notice. Granting access also lets Citi share account information with the tool, and the bank's privacy notice does not cover what the tool does with it.
Some AI companies are working around the problem rather than fighting it. The personal finance tool Anthropic launched this month is billed as read-only and says it cannot move, manage or invest money. That avoids the liability question by removing the ability to transact, at the cost of the features that make an agent useful.
For readers, the practical question is what their own bank says. Nothing in the Citi update changes the protection for fraud a customer did not authorize, but it does draw a line at anything done through a tool the customer connected. Regulators have not published guidance on how these clauses sit against consumer protection rules, and none of the banks has said how many customers use AI agents today. TechcityAuthority covered OpenAI's move to let agents complete purchases earlier this month, the same shift from advice to action that banks are now pricing in.

Comments