JPMorgan Chase and Charles Schwab are blocking outside AI assistants that try to enter customers’ accounts with usernames and passwords, even when the customer supplied the credentials. JPMorgan described the practice on its October 13 earnings call; Schwab’s chief executive separately said the firm already blocks agents trying to log in for clients.
That creates two separate decisions. A customer may authorize an assistant to help with money, while the bank prohibits the way it signs in. If the assistant then moves money incorrectly, responsibility may depend on the authority the customer granted, the payment method and the consumer protections that apply.
A customer can ask an assistant to help. The bank can still reject password-based access or impose conditions on how the assistant connects. Separately, whether a disputed transfer qualifies for reimbursement depends on the authority granted and the law covering that transaction. The bank’s description of an agent as unauthorized does not decide that question by itself.
JPMorgan chief financial officer Jeremy Barnum said some services that let agents move money rely on credential sharing, and that the bank blocks them when it can. Chase’s current terms draw a narrower line than a blanket ban: agents must identify themselves, label their traffic, answer identity checks truthfully and stop when requested. Chase can still restrict or terminate their access.
Axios reported a broader response involving Morgan Stanley, Bank of America, JPMorgan, Wells Fargo and other banks, citing people familiar with the issue. Its unnamed sources also said agents had contacted call centers or appeared to attempt actions independently of users’ instructions. The report does not identify an agent provider, requested task, attempted transaction, dollar amount or customer loss.
Morgan Stanley told Axios it had not observed “rogue AI agents accessing our platforms.” The article did not include on-record confirmation of a specific incident from Bank of America or Wells Fargo. Nor does the published reporting identify a customer whose account was frozen or closed, or whose own online access was disabled, after an agent tried to enter. The confirmed response is a blocked login attempt, not a frozen customer account.
The banks’ written rules show how password sharing, agent access and responsibility can diverge. Timing matters: Wells Fargo’s AI-specific provisions take effect on November 24 and are not the rules in force on October 22.
| Bank and timing | What the documented rules say | What customers should understand |
|---|---|---|
| Chase: current terms | Agents must identify themselves, label their traffic, answer identity checks truthfully and stop when requested. Chase can restrict or terminate their access. | Meeting those conditions does not guarantee access. The terms assign an agent’s interpretation errors, unintended actions and actions beyond its authority to the user, subject to applicable law and other agreements. |
| Bank of America: effective October 19 | The agreement makes customers responsible for transactions by someone permitted to use their credentials, even when that person exceeds the authority granted, until the bank is notified and can block and replace the credentials. | The clause refers to another person and does not expressly address AI software or its provider. Covered consumer transfers remain governed by Regulation E. |
| Wells Fargo: current and forthcoming | The May 21 agreement has general credential and access provisions. The November 24 version adds explicit responsibility for AI-agent instructions, unintended actions and actions beyond the authority granted. | The forthcoming rules require customers to monitor agents and retain evidence that authorization was revoked, while preserving protections required by law. They do not establish that Wells Fargo offers an approved consumer-agent connection. |
Wells Fargo’s pending language makes the practical burden especially explicit. Customers must stop an agent’s activities, retain evidence that permission was revoked and notify the bank. To the extent allowed by law, responsibility continues until that notification. The agreement also says Wells Fargo’s privacy and security policies do not govern how an outside agent or its provider handles information it receives.
Bank contracts do not settle every loss. Regulation E covers qualifying electronic transfers from consumer accounts. Its official interpretation says that when a consumer gives another person an access device and authority to make transfers, the consumer is fully liable if that person exceeds the authority granted, until the bank is notified that the authority has ended.
The CFPB materials do not say how an AI model, software agent or agent provider fits that rule. Permission to read balances also does not necessarily include permission to transfer money. The actual authority granted would matter.
Credentials obtained through fraud or robbery are treated differently. CFPB guidance describes investigation duties and liability protections for covered unauthorized transfers. Those rules do not apply wholesale to business accounts or every payment method, but sharing access does not automatically erase every consumer protection.
The practical rule is to treat a request for a bank password as a request for financial authority, not a routine setup step. Use a bank-approved connection that limits what an assistant can see or do and can be revoked. If the bank offers no approved route, the customer’s permission does not create one. If disputed activity appears, report it promptly.