Wells Fargo wants customers to show that they stopped AI agents

Wells Fargo, a US bank, changed its online access agreement. The terms start on 24 November. A customer must cancel the permission of an AI agent and keep a record of it. The customer gives the record to the bank if the bank tells the customer to do so. Until the customer tells the bank, the customer has the risk of all that the agent does, as far as the law permits. This appears to be a first at a bank.

What changes

JPMorganChase and Wells Fargo tell customers that they have the risk of AI agents they authorize, also for tasks that nobody told the agent to do. Wells Fargo adds a duty to keep a record. A spokesman tells that the bank possibly cannot see when a customer cuts off an agent.

The proof

The bank gave no data on agent incidents. Its reply did not tell what counts as a record. Douglas Borthwick, a past Morgan Stanley executive, tells that a screenshot shows only that the customer told the agent to stop. Screenshots are easy to fake.

What is not known

Elaine Duffus of Wolters Kluwer tells of an open question. Who is at fault is not known if an agent does a task that no one told it to do, after the customer cancelled it with a record. She tells that banks must have agent governance to see problems and stop an agent.

Insight: Banks block agents or move the risk to the customer

Citigroup does not let external agents use accounts. Bank of America wants a human to agree before money moves. Wells Fargo lets agents in, but puts the proof on the customer. Torsten Slok of Apollo tells that agents can move cash from accounts that pay 0.1% to accounts that pay 3.3% to 5%. This reads as a toll gate: the customer must keep the proof, and that possibly slows agents that move cash.