Skip to content
Thursday, November 19, 2026Free guide
EN
dailyai
Features

Citi Treats AI Agent Bank Transfers as the Customer’s Own

Citibank's November update treats transfers started by a third-party AI agent as the customer's own, even when they conflict with the customer's prompts. Other large US banks are writing similar rules, though federal electronic-transfer rights cannot be waived by contract.

· 5 min read
ShareXin

Citibank’s November 2026 online banking terms treat a transfer started by a third-party AI agent as one the customer made, even when it clashes with the customer’s prompts.

The November online user agreement adds a new section on third-party AI agents. It cautions customers against using such software to reach Citi Online or the Citi Mobile app. If they do so anyway, the bank says it will treat the agent’s access and actions as authorised.

That stance also covers a later leak. If the login details given to the agent are then shared, leaked or passed to other people or other agents, the bank says it will still treat any access or action that uses those details as authorised by the customer. It says it has no duty to check whether those actions were valid, even when they conflict with the prompts the customer gave the software.

The bank defines an AI agent broadly as software that can take action on a customer’s behalf, whether it acts on its own, with some independence, or only when instructed. That is wider than a chatbot that only answers questions. It includes tools that can log into an online bank, read transactions or start a payment.

What the November terms say

The funds-transfer terms say the customer is solely responsible for instructions an agent sends. Any transfer started by the agent is treated as started and authorised by the customer, without further investigation. The November notice on that page points readers to the same rules in the online user agreement and the mobile terms of use.

The new section does not apply to AI tools that the bank itself provides through those channels. The bank also reserves the right to limit, suspend or block a third-party agent at any time, without notice.

By giving an agent access, the customer authorises the bank to pass information, including sensitive personal information, to that software. The bank’s privacy notice does not cover how the outside agent collects, stores, protects or shares that data. Monitoring the agent, the agreement says, is the customer’s job, not the bank’s.

There is a legal caveat in the same section. The customer is responsible for the agent’s errors, for unintended actions, and for actions that go beyond the authority they granted, “except as otherwise permitted under applicable law.”

An 18 November report quoted a bank spokesperson on the reason for the update. The spokesperson said the bank wanted customers to understand the risks of sharing banking credentials or authorising third-party AI agents, and to “reinforce that customers who choose to grant account access to third-party AI agents are responsible for activities conducted through those agents.” The report described Citibank as the third-largest commercial bank in the United States.

Want AI to save you hours a week? Our free playbook shows you how.Get the free guide →

Other banks are writing similar rules

Citibank is not acting alone. A customer notice says the changes take effect on 24 November 2026. The summary of Wells Fargo’s updated Online Access Agreement says that if a customer uses a third-party AI tool or agent to reach their accounts, requests submitted through that tool may be treated as if they came directly from the customer. The customer is responsible for managing the tool’s access. The bank also reserves the right to restrict or block certain tools.

The upcoming online access agreement says agent transactions are treated as the customer’s, unless the law says otherwise. If a customer gives an agent their login details, or otherwise authorises it to act, instructions, actions and transactions started by the agent are “deemed to have been initiated and authorized directly by you, without further verification.” The bank says it has no duty to check the agent’s authority. Use of such an agent is “at your own risk.” The bank also says its privacy and security policies do not apply to the outside agent, and that it does not monitor the agent’s activity.

A 27 October report described a customer email, later confirmed by a bank spokesperson, warning people not to share login details with an AI agent. Bank of America told customers not to give their username and password to an “AI-enabled tool or agent that offers to act on your behalf.” The email added: “If you do share your credentials, keep in mind that you’re responsible for activity resulting from that access.”

The 18 November report also said JPMorgan had announced similar policies in October, that PNC changed its terms that day, and that Citizens emailed customers the same week with a comparable warning. Those three updates were not independently reviewed for this article.

The same report noted that a personal finance tool from Anthropic, launched earlier in November, is billed as read-only and “can’t move, manage, or invest your money.” A DailyAI report described a payments wallet designed so an agent can pay without seeing the customer’s card number. Banks are now drawing a sharp line around a different kind of access: handing an outside assistant the keys to the bank account itself.

What the federal rules still protect

The bank contracts are not the last word for ordinary consumer accounts. In the United States, the Electronic Fund Transfer Act and its rulebook, Regulation E, limit how much a customer can be held liable for an unauthorised electronic transfer. They also set out how a bank must investigate when a customer reports an error.

Federal electronic-transfer guidance says a bank contract cannot waive those rights. One question in the Consumer Financial Protection Bureau’s FAQs asks whether a bank can rely on an agreement that drops Regulation E protections if the customer has shared account information with a third party. The answer is no. The statute says no agreement may waive a right the law confers. A contract that tries to do so, the bureau says, violates the Act.

The same guidance says a transfer can still be unauthorised when a fraudster uses login details obtained by hacking a third party the customer had shared them with, if the customer did not actually authorise that transfer and did not benefit from it. That is close to the leaked-credentials case the November bank terms describe as authorised by the customer. The federal guidance points the other way when the later use is fraud.

The harder case is a mistake by an agent the customer did mean to use: a payment to the wrong account, or an action that goes beyond the prompts. Citibank says it will still treat those as authorised, subject to applicable law. Whether any given transfer is unauthorised under Regulation E depends on the facts, including whether the customer gave actual authority. The federal rules also depend on how quickly a problem is reported. Customers who see a bad transfer should still tell their bank promptly, rather than assume the contract has closed the matter.

Why it matters: AI assistants are starting to reach real bank accounts, not just chat windows. Large US banks are answering by putting much of the contractual risk of that access onto the customer, including in some terms when login details later leak. Those contracts cannot wipe out federal rights against unauthorised transfers, but they do show how banks intend to treat disputes. Anyone who connects an outside assistant to a bank account should read the terms, prefer tools that cannot move money, and report a bad transfer quickly.