Wells Fargo shifts AI agent liability onto customers

Wells Fargo has quietly redrawn the legal boundary between a bank customer and the software acting on their behalf. An updated Online Access Agreement, taking effect on November 24, 2026, adds a new Section 15 covering third-party AI agents. The document, reviewed by American Banker, states that anything an agent does under a customer's login is deemed to be initiated and authorized directly by you, without additional verification.

Proof of revocation now sits with the customer

The clause goes further than most digital banking terms ever have. Customers must retain proof of revocation of an agent authorization and provide it to the bank upon request. In practice, if you delegate an AI assistant to pay bills, move funds or check balances and later decide to cut it off, the burden of demonstrating that revocation falls on you, not the bank.

Liability also applies before the bank is ever notified. Until Wells Fargo is informed, the customer carries responsibility for the agent actions, explicitly including unintentional actions and the consequences of manipulated inputs — language that covers prompt injection, where a malicious string hidden in a web page or an email tricks an agent into a harmful transaction. The bank frames the entire category as use AT YOUR OWN RISK.

First documented shift of its kind at a major bank

This appears to be the first documented liability transfer of this kind at a major bank. It lands amid a broader industry retrenchment — a wave of bank blocks first reported by Axios and Reuters on October 22, with JPMorgan, Bank of America, Citi and Morgan Stanley blocking third-party agents from customer accounts. Where peers built walls, Wells Fargo built a waiver.

The timing is not accidental. Agentic browsers and personal AI assistants that log into websites on behalf of a user went mainstream this year, and banks must decide whether such access is fraud, an unauthorized login, or a new product surface. Wells Fargo chose a third path — allow it, but place the risk on the only party that controls the agent, the account holder.

IMPACT: If you run AI agents against your bank login, expect the default rule to become simple — you own every action the agent takes. Other banks will likely copy this template, so review agent permissions before November 24, 2026 and keep written proof whenever you revoke access.

Continue the conversation: https://t.me/AIAutomationArchitecture · Bluesky