Banks tighten AI rules
Citi updated its online banking terms in November to make customers responsible for the actions of third-party AI agents, even if an agent leaks credentials or ignores instructions. This revision clarifies that Citi has no obligation to investigate whether an agent followed a customer’s original prompts, and the provision does not apply to AI agents offered by Citi itself. I find this policy shift concerning because it leaves users vulnerable to software that acts without permission. JPMorgan and Bank of America announced similar policies in October, while PNC updated its digital agreement on Wednesday. Citizens Bank warned customers via email this week that they could be responsible for what their AI assistants do. PNC distinguishes between "bank-approved connections" and handing an agent unrestricted access to an account, and its rules state that customers cannot give third-party AI agents their passwords or "one-time security codes," nor can agents initiate payments unless PNC expressly permits it.
Federal rules limit bank power
The industry move follows a security scare on October 22 when major banks detected AI agents logging into customer accounts and taking steps that appeared to go beyond user instructions. An AI agent differs from a conventional chatbot because it can act on a user’s behalf through tasks such as navigating websites, accessing financial records, or initiating payments. This capability makes a compromised login much more dangerous than a simple chatbot error.
| Bank | AI Policy Status |
|---|---|
| Citi | Updated terms in November |
| JPMorgan | Policy announced in October |
| Bank of America | Policy announced in October |
| PNC | Updated agreement Wednesday |
| Citizens | Warned customers this week |
Federal consumer protection rules generally require banks to investigate unauthorized electronic transfers and limit customer liability for fraud, and these protections cannot be waived through a banking agreement. The Consumer Financial Protection Bureau states that banks cannot use agreements to waive those legal protections, although the application of these rules to transactions performed by customer-authorized AI agents remains unclear. I think the banks are attempting to exploit the distinction between theft by an outsider and transfers actually authorized by a customer or someone given access. Whether a particular loss qualifies as unauthorized depends on how the transaction occurred. Because federal rules still restrict when banks can hold consumers liable for genuinely unauthorized electronic transfers, the bank’s new language does not automatically eliminate consumer fraud protections. The Office of the Comptroller of the Currency has since issued AI examiner guidance that addresses outside agents, payment controls, and human approval, but the framework is nonbinding and does not itself create new requirements for banks.
