Wells Fargo and Chase Terms Put AI-Agent Mistakes on Customers, Not Banks
Wells Fargo's updated Online Access Agreement, effective November 24, says requests submitted through a third-party AI tool or agent may be treated as coming directly from the customer, and reserves the right to block certain tools. JPMorgan Chase's Digital Services Agreement, effective October 27, bars outside AI from using customers' login credentials, permits approved APIs, and deems customers to authorize an outside AI's actions if they let it in. WIRED reports Bank of America, Citibank and PNC have also added agent warnings; we read only the Wells Fargo and Chase texts.

The banks' answer to AI agents is not a ban but an allocation of blame. From Tuesday, November 24, Wells Fargo's updated Online Access Agreement says that requests submitted through a third-party AI tool or agent "may be treated as if they came directly from you," and reserves the right to restrict or block certain tools. JPMorgan Chase's Digital Services Agreement has said since October 27 that a customer who lets an outside AI reach their accounts is deemed to have authorized whatever that AI does.
Both documents leave a door open. Chase permits outside tools, including outside AI, to read account data through approved APIs with the customer's authorization; what it prohibits is using a customer's login credentials. Our read: the clauses are less about stopping agents than about settling who bears the first loss when one misfires, and the customer comes first in both.
What the agreements say#
Wells Fargo summarizes the change on its customer notice. If a customer uses a third-party AI tool or agent to reach their accounts or services, the bank says, the tool is not operated by Wells Fargo and the customer authorizes the bank to share the information the service needs. That information is then subject to the tool provider's privacy practices and terms. Requests submitted through the tool "may be treated as if they came directly from you," and the agreement sets out the customer's responsibilities for managing the tool's access. The bank also reserves the right to restrict or block access from certain third-party AI tools. The notice cites Sections 15, 17(c) and 20(c) of the agreement, which applies to anyone who keeps using the services after November 24.
Chase's text is more detailed. Third parties, "including External AI, automated tools, or data aggregators," are prohibited from using a customer's login credentials, and customers who shared credentials must change them. A customer who permits, directs or facilitates a third party's access is deemed to authorize that party to transact in the account and to accept that the bank may treat its instructions as the customer's own. The customer also acknowledges that such actions "may be inaccurate, incomplete, unintended, or inconsistent with your actual requests," and agrees to be responsible for resulting errors, including unintended transactions. Chase may deny, suspend or terminate access, including access facilitated by outside AI, "in our sole discretion and without notice," and the indemnity clause covers the conduct of any agent, including External AI, "whether or not such conduct is intended by you." The agreement defines External AI as third-party AI and third-party AI agents.
What WIRED adds#
WIRED, in a November 22 piece by Reece Rogers, quotes the Wells Fargo terms as saying: "Your use of any AI Agent is AT YOUR OWN RISK." We found that wording only in WIRED; it is not on the bank's summary page, and we did not read the full agreement. WIRED also reports that Bank of America warned customers the same day as Chase's change not to share login details with agents, and that Citibank and PNC have changed their terms to include agent warnings. We did not read those texts.
WIRED's most useful framing comes from Luca Cian, a University of Virginia professor who studies consumer psychology and AI: "Banks are doing two things at once. They're locking the door to outside agents and building their own." WIRED cites Wells Fargo's own generative AI assistant, which it says users have interacted with more than a billion times. It also reports that American Express has announced that some agent mistakes, such as ordering too much of an item, may become eligible for reimbursement if the money cannot be recovered from the retailer. That is the only sign in the reporting that any financial firm is moving the loss the other way.
Where the loss lands#
The customer-first rule is not new in the market. On its October 13 earnings call, Jamie Dimon told investors there is "a liability transfer" when customers use outside agents, and CFO Jeremy Barnum said that if an agent does something a customer does not want, "that's going to be on them," according to a transcript. In Congress, a discussion draft from Rep. Lori Trahan would instead make AI developers liable when their systems act in ways that would be negligent or criminal if a person did them; it is a draft, not law.
The bank terms we read settle only the customer-to-bank relationship. They say nothing, in the passages we read, about what an agent's developer or a merchant owes the customer.
Limits#
- Wells Fargo's terms are read from the bank's customer notice, which summarizes the changes; we did not read the full agreement, and the "AT YOUR OWN RISK" wording is WIRED's quotation.
- Chase's text is from its Digital Services Agreement PDF; we read the change notice and the sections cited above, not the whole document. The deemed-authorization language applies to customers who permit, direct or facilitate a third party's access.
- The Bank of America, Citibank and PNC changes, and American Express's reimbursement plan, are as reported by WIRED.
- These are contract terms. We found nothing in the documents we read on how they interact with consumer-protection rules for unauthorized transactions.
- Chase's change took effect before this edition's window; we include it because Wells Fargo's takes effect on Tuesday and WIRED set the two side by side on November 22.