Banks warn AI agents are your problem when they go wrong
US banks are warning customers that third-party AI agents can make costly mistakes — and the bill may land with the account holder.
Wells Fargo is warning millions of customers that using third-party AI agents to access online banking is entirely at their own risk. The bank’s updated agreement says customers are responsible for what happens when an automated tool logs in, clicks through a service or makes a transaction on their behalf.2
The warning lands as “agentic” software moves from chat windows into shopping, travel and account-management tasks. These tools can pursue a goal — such as buying an item or booking a service — rather than waiting for a person to approve every click. That convenience is also the problem: an agent can misunderstand an instruction, expose credentials or carry out a purchase the user did not intend.
Why banks are tightening the rules
Wells Fargo is not alone. WIRED reports that JPMorgan Chase changed its terms in October to ban third-party agents that log in with a customer’s credentials, while Bank of America warned customers not to share login information with agents.134 Citibank and PNC have also added warnings or changed their policies, according to the report.1
The banks’ position is straightforward. If a customer authorises an outside bot, the bank does not want to be treated as responsible for the bot’s mistakes. A faulty agent might place an unwanted order, trigger an automated withdrawal or expose sensitive account information before anyone notices.
That risk is still emerging rather than universal. WIRED notes that the share of online transactions completed by autonomous tools remains small, even as early adopters use them for shopping and other repetitive jobs.1 The same report also describes banks developing their own first-party AI tools while restricting outside agents — a sign that the argument is partly about control as well as security.
The distinction matters. Banks are not rejecting automation altogether; they are drawing a line around software they do not operate, audit or control. American Express, for example, has discussed potential reimbursement for some mistakes made during agentic purchases, but that is a limited protection rather than a general safety net.5
What this means for customers
The practical lesson is to treat an AI agent with access to a bank account as an assistant with payment authority, not as a harmless browser extension. Check the permissions it requests, avoid handing over login details unless the service is explicitly supported, and review account activity after allowing any automated purchase or transfer. The exact rules will depend on the bank and the country, so customers should read the current agreement for their own provider rather than assume another bank’s policy applies.
For UK readers, the Wells Fargo change is a US example rather than a change to British banking rules. But the underlying question — who carries the loss when an AI tool makes a costly mistake? — is likely to matter wherever agentic software can act inside financial services.
Our opinion
Banks are right to make the risk visible before agentic software becomes normalised. “Your use of any AI Agent is AT YOUR OWN RISK” is blunt, but customers should not mistake a polished interface for human judgement. The weak point is accountability: banks can reject outside bots, AI companies can point to disclaimers, and retailers can say an order was validly placed. Until those responsibilities are clearer, giving an agent unrestricted access to a bank account is a poor trade for convenience.
The broader story is not that AI agents are useless. It is that they need narrower permissions, better audit trails and a clear way to undo mistakes before banks can reasonably expect ordinary customers to trust them with money.