Citi Holds Customers Liable for Third-Party AI Agent Actions, Even After Credential Leaks
Summary: Citi’s updated online banking terms treat actions by third-party AI agents as customer-authorized, even if credentials are leaked or the agent ignores instructions. The move follows similar policy tightening by JPMorgan, Bank of America, and PNC after a security scare in October.
Citibank has updated its online banking terms to hold customers responsible for the actions of third-party artificial intelligence agents accessing their accounts, including scenarios where login credentials are leaked or hacked. The November revision explicitly treats any activity conducted by an outside AI tool as authorized by the customer, regardless of whether the software ignores specific instructions or exceeds its granted permissions.
Under Citi’s new terms, if a user provides their online credentials to a third-party AI agent and those credentials are subsequently shared, leaked, or distributed to other parties, Citi will consider any resulting access or transactions as authorized. The bank states it has no obligation to investigate whether the agent followed the customer's original prompts, subject to applicable law. A Citi spokesperson told TechCrunch that the update is intended to clarify the risks associated with sharing banking credentials or granting external tools account access, noting that the provision does not apply to AI agents offered directly by Citi.
The policy shift addresses a growing challenge for financial institutions as AI agents, which can act autonomously on a user’s behalf to navigate websites and initiate payments, become more prevalent. Unlike conventional chatbots, these agents can execute financial transactions, making mistaken instructions or compromised logins significantly more consequential. The change comes in the wake of a banking security scare from October 22, during which major banks detected AI agents logging into customer accounts and taking steps that appeared to go beyond user instructions. Executives reported struggling to determine which agents had been properly authorized.
Citi is not acting in isolation. JPMorgan and Bank of America announced similar policies in October, while Citizens Bank warned customers via email this week that they could be held responsible for the actions of their AI assistants. PNC also updated its digital agreement on Wednesday, taking a stricter operational approach. PNC’s rules distinguish between bank-approved connections, which provide limited access, and unrestricted access granted by handing over passwords. Under PNC’s updated terms, customers are prohibited from giving third-party AI agents their passwords or one-time security codes, and agents generally cannot initiate payments unless expressly permitted by the bank.
Despite these contractual changes, federal regulations may still limit how much liability banks can shift to consumers. Under U.S. federal consumer protection rules, specifically the Electronic Fund Transfer Act and Regulation E, banks are generally required to investigate unauthorized electronic transfers and limit customers' financial liability for fraud. The Consumer Financial Protection Bureau states that these legal protections cannot simply be waived through a banking agreement. However, the application of these rules to transactions performed by customer-authorized AI agents remains less clear, as regulators distinguish between theft by an outsider and transfers actually authorized by a customer or someone given access.
The Office of the Comptroller of the Currency has issued nonbinding AI examiner guidance addressing outside agents, payment controls, and human approval, but this framework does not create new legal requirements for banks. As AI startups increasingly target personal finance to analyze spending and manage bills, the tension between innovation and security continues to shape banking policies. While banks view unauthorized AI access as a competitive and security threat, the legal boundary between customer authorization and fraudulent exploitation remains a critical point of contention.