-
Key insight:Wells Fargo appears to be the first bank to start requiring customers to prove they've turned off their AI agents' access to their bank accounts, or else be liable for any rogue AI agent activity.
-
Expert quote:"What if I do everything Wells Fargo says to do, and I've got a pile of evidence that I told that agent to stop doing that thing, yet the agent went rogue because maybe there was an ambiguous direction to it or it just went bad, there's still that question — what happens then?"—Elaine Duffus, senior specialized consultant at Wolters Kluwer.
Big banks including JPMorganChase and Wells Fargo have been changing their terms of conditions for online and digital banking users to make it clear that customers are liable for any actions taken by AI agents they've authorized — even when those agents go rogue and do something they were not prompted or authorized to do.
Wells Fargo is taking this concept a step farther. An updated version of its online access agreement that takes effect Nov. 24 states, "You are responsible for properly and timely revoking any authorization provided to an AI Agent and taking any necessary further action to cease the AI Agent's activities. You must retain evidence of such revocation and must provide such evidence to us, at our request. Until you notify us of such revocation pursuant to Section 17(d), you remain responsible for all the AI Agent's activities, to the extent permitted by law or regulation."
In other words, even if a customer tells an AI agent to stop accessing their bank account, they're still on the hook unless they produce evidence that they revoked authorization.
In response to several questions about the new policy, including what counts as evidence of revocation, a Wells Fargo spokesman said, "Customers may choose to authorize or revoke an AI agent outside of Wells Fargo. The updated agreement asks customers to keep a record of when an AI agent's authority was revoked, because Wells Fargo may not otherwise have visibility into when that occurred."
This raises questions of whether it's reasonable to expect consumers to capture such evidence and what banks would accept as evidence.
"It's fair for Wells Fargo to want a record of when an agent was cut off, since the bank can't see it," said Douglas Borthwick, a former Morgan Stanley and Standard Chartered executive who hosts the Old Men, New Money podcast. "It isn't realistic to expect customers to produce one. Telling a chatbot to stop leaves nothing in the customer's hands that anyone else can check."
A screenshot of the chat in which the customer revoked access "shows you asked. It doesn't show the agent stopped, and screenshots are easy to fake." Borthwick is the founder of a startup, The Insumer Model, that provides wallet authentication for developers and AI agents.
Even if a customer revokes her AI agent's access to her bank account and can prove she did so, the consumer may not be covered if the agent goes off script.
"What if I do everything Wells Fargo says to do, and I've got a pile of evidence that I told that agent to stop doing that thing, yet the agent went rogue because maybe there was an ambiguous direction to it or it just went bad, there's still that question — what happens then?" Elaine Duffus, senior specialized consultant at Wolters Kluwer, told American Banker. "Who's at fault then? Is it me? Is it the bank? Is it the AI provider? How far do we go down the chain to figure out maybe it was that company's fault or that person's fault that this thing had a mind of its own and went and did something that it wasn't supposed to do."
This all calls for a governance process or agent-governance layer, Duffus said, which is something Wolters Kluwer is developing, "so that if things really start to go off the rails, the bank will see it and then can reach out to the customer or just hit the kill switch."
