US banks tighten rules on third-party AI agents that act on customer accounts
Wells Fargo, JPMorgan Chase, Bank of America, Citibank and PNC have issued new terms or warnings about AI agents, even as they build their own tools, WIRED reported.
· 2 min read
- Wells Fargo's updated customer agreement says use of any AI agent is at the customer's own risk.
- JPMorgan Chase, Bank of America, Citibank and PNC have also changed terms or issued warnings about agents.
- American Express said some agent errors may be eligible for reimbursement in the future.
Several large US banks are moving to limit how third-party AI agents can interact with customer accounts, according to WIRED. Wells Fargo is set to put an updated customer agreement into effect on Tuesday, and the document places responsibility for agent use squarely on customers: "Your use of any AI Agent is AT YOUR OWN RISK."
AI agents are tools that automate online tasks and click on a user's behalf. WIRED described them as still nascent after a bumpy 2026. A growing group of power users has adopted them, with mixed results. Some benefit from the automation, while others have made embarrassing errors, such as one case in which an agent posted bank details where coworkers could see them.
Wells Fargo is not alone. WIRED reported that in late October JPMorgan Chase changed its terms to ban third-party agents that log in with a customer's credentials. On the same day, Bank of America warned customers not to share login information with agents. Citibank and PNC have also updated their terms to include warnings about agents.
The friction extends beyond finance. Amazon has blocked Meta's Muse, one of the most widely used agents, from making purchases on its platform, and United recently said personal bots would be barred from buying flights on a traveler's behalf.
Experts quoted by WIRED see the bank policies as precautionary. Sucharita Kodali, a Forrester analyst, said most people do not even know what agents are, much less use them, and described banks as risk-averse institutions that worry about transaction security and automated withdrawals that could harm account holders. Kacy Kim of Bryant University said consumers authorize goals, not every click, and she expects some to find it unfair to be held accountable for an agent's mistakes. She does not expect the new rules to deter dedicated early adopters.
Luca Cian, a University of Virginia professor, said shoppers are comfortable automating routine purchases such as household supplies but are more wary of handing over big-ticket buys. He offered a hypothetical in which an agent accidentally orders 10 roasted chickens, raising the question of whether the retailer, the AI developer or the bank bears the blame. Under the new bank terms, the bank is making clear it does not want that role.
Some card providers are exploring limited protection. American Express announced that certain agent errors, such as ordering too many of an item, may become eligible for reimbursement through the card provider if the money cannot be recovered from the retailer.
At the same time, banks are investing in their own AI. Cian said they are locking the door to outside agents while building their own. WIRED noted that users have interacted with Wells Fargo's generative AI assistant more than a billion times, Bank of America revamped its EricaAssist agent to help employees this year, and Citigroup launched an internal development platform centered on agents.
Kodali questioned whether agentic shopping solves a real problem, noting that buying from large retailers is already easy. For now, WIRED concluded, the banks' position is that the agent is the customer's responsibility.