Citi has added AI-agent rules to its online banking agreement that put responsibility for an outside assistant's actions on the customer who gives it access. Wells Fargo has published a similar policy that takes effect November 24.

For customers considering an AI tool to handle banking, the immediate decision is what account access to grant. Both banks say they can treat an agent's instructions as the customer's own, including actions that exceed the customer's directions. Federal protections for covered unauthorized transfers still apply.

What Citi's November update says

Citi's Online User Agreement identifies its new Section H as a November 2026 addition. It cautions customers against outside AI agents accessing Citi Online or Citi Mobile. The definition includes tools taking autonomous, partly autonomous or non-autonomous action for a customer.

If customers provide credentials or otherwise allow access, Citi says it will consider the agent's actions authorized. The clause also covers credentials subsequently shared or leaked to other people or agents. It places monitoring responsibility on the customer and says Citi's privacy notice does not govern the outside tool. Citi-provided AI features are excluded from these particular terms.

Citi also updated its separate funds-transfer agreement. Section 4 says it treats instructions from an agent acting for a customer as authorized even when they conflict with that customer's prompts, without investigating the scope of the agent's authority. That agreement covers requests from U.S. Citibank accounts, including wire transfers and other specified transfers.

Wells Fargo adds a revocation requirement

Wells Fargo bank building with red signs in downtown Cheyenne
Wells Fargo in downtown Cheyenne, Wyoming, photographed June 29, 2024.Tony Webster / Wikimedia Commons, CC BY 2.0. Reduced resolution.

The upcoming Wells Fargo agreement, marked effective November 24, covers its online, mobile, business and advisor services. Its AI-agent section treats instructions as customer-authorized unless law or regulation provides otherwise, and says the bank does not monitor or certify an outside agent's activity.

It also tells customers how responsibility ends: revoke the agent's authority, take the steps needed to stop its activity, retain evidence of that revocation and provide it if the bank asks. Responsibility continues until the customer notifies Wells Fargo, to the extent permitted by law or regulation. Section 17(d) provides a 24-hour notification number, 800-956-4442.

Telling the assistant to stop and informing the bank are separate actions under those terms. Keep the revocation confirmation and the bank notification record. Wells Fargo also reserves the ability to restrict or terminate an agent's access; the agreement does not promise every tool will be allowed.

Federal protections for unauthorized transfers

Glass entrance of the Consumer Financial Protection Bureau headquarters in Washington
The Consumer Financial Protection Bureau headquarters entrance in Washington, photographed February 10, 2025.G. Edward Johnson / Wikimedia Commons, CC BY 4.0. Reduced resolution.

The Consumer Financial Protection Bureau's transfer guidance says an agreement cannot waive rights granted by the Electronic Fund Transfer Act. Its examples include a hacker obtaining credentials from a third party and using them to make a covered transfer without the customer's actual authority and without benefiting the customer. Such a transfer is unauthorized under Regulation E, the bureau says.

That guidance gives customers a reason to report a disputed transfer even when a bank's AI clause describes broad responsibility. It does not decide how every AI-agent dispute will turn out.

Authorization also has limits of its own. The regulation's definition and commentary distinguish a person given an access device and transfer authority from someone who obtains access through fraud. The commentary says someone who grants another person transfer authority can remain liable when that person exceeds it, until the financial institution is notified that authority has ended. Applying those rules requires the facts of the transaction.

Payment type matters, too. Regulation E's coverage exclusions include Fedwire and similar wire systems used primarily between financial institutions or businesses. The rules for a covered consumer electronic transfer should not be assumed to govern every wire or investment transaction.

If an agent makes a disputed transfer

Notify the bank promptly through its official error-reporting channel and identify the transaction, amount, date and reason for the dispute. Keep the agent's instructions, activity history and any revocation evidence so the bank can distinguish what you authorized from what occurred.

For covered errors, Regulation E generally requires notice within 60 days after the bank sends the statement first showing the error. An oral report can trigger an investigation, although the bank may require written confirmation within 10 business days. Reporting lost or stolen access credentials promptly also matters: liability rules can depend on whether notice arrives within two business days after learning of the loss or theft.