Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarWired→ original

Wells Fargo and four other US banks stop taking responsibility for customers' AI agent mistakes

Starting Tuesday, Wells Fargo is adding a clause for millions of customers: any AI agent is used "at your own risk." Four other major US banks have already taken a similar step. JPMorgan banned agents that log in with a customer's password, while Bank of America, Citibank and PNC added warnings. If an agent mistakenly orders ten fried chickens, the bank won't pay for it. Developers of payment agents need spending limits and human confirmation.

AI-processed from Wired; edited by Hamidun News
Wells Fargo and four other US banks stop taking responsibility for customers' AI agent mistakes
Source: Wired. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceWired ↗
Editor-in-Chief
Key metric #1Amazon blocked Meta's Muse agent, and United closed its ticket purchases to bots
Key metric #2American Express promises to reimburse some agent errors, such as an extra order
Key metric #3Wells Fargo's generative AI assistant has been used more than a billion times

Imagine you asked an AI agent to order groceries, and it bought ten fried chickens. Who pays? The store will say it processed a normal order.

The agent's developer will point to the disclaimers about errors in the terms. Now banks are saying they don't intend to take responsibility either. Starting Tuesday, Wells Fargo is rolling out an updated agreement for millions of customers that states directly: using any AI agent happens "at your own risk."

This is the fifth major US bank to adopt this position. Before it, JPMorgan Chase, Bank of America, Citibank and PNC changed their terms or issued warnings. Here, an "AI agent" means a bot that clicks through websites on its own and takes actions on your behalf, such as making purchases.

What changed in the banks' terms

Wells Fargo is not the first. In late October, JPMorgan Chase changed its terms and banned third-party agents that log into an account using the customer's login and password. That same day, Bank of America warned customers not to give agents their login credentials. Citibank and PNC also added warnings about agents to their terms.

The logic is simple. An agent that clicks for you looks to the bank like the customer himself. If it made a mistake, and you didn't manually confirm anything, it's hard to prove whose fault it was. As Casey Kim, a marketing professor at Bryant University, put it, consumers authorize goals, not every click. Banks don't want to be responsible for the consequences of those "goals."

Not just banks: a fight for territory

Banks are not alone. Amazon blocked Meta's Muse agent, one of the most popular, and does not let it make purchases on its platform. The airline United announced that personal bots will not be able to buy tickets on a user's behalf. Websites and banks want to see a live customer, or at least their own verified intermediary.

There are still few real users of agents. Forrester analyst Sucharita Kodali says most people don't even know what agents are, let alone use them. In her view, banks are cautious by nature and prefer to hedge in advance, even if incidents remain rare. Among the risks she names transaction security and automatic debits that could harm the account holder.

University of Virginia professor Luca Chan adds a behavioral detail: people are willing to trust AI with buying toilet paper, that is, routine expenses. Expensive and non-standard purchases, where caution is needed, are not being rushed toward automation.

Your own agents are fine, others' are not

Banks are not rejecting agents altogether. According to Chan, they are doing two things at once: closing the door to external agents and building their own. Users have turned to Wells Fargo's generative AI assistant more than a billion times. Bank of America updated its EricaAssist agent this year to help employees. Citigroup launched an internal platform for developing agents.

There are also first attempts to soften the blow for customers. American Express announced that some agent errors, such as ordering too much of a product, may in the future be compensated if the money could not be recovered from the seller. For now, this is a promise, not a finished program with terms.

What this means for those building agents

First, the scheme of "the agent logs into the user's account with their login and password" is becoming legally and technically vulnerable: banks are banning it or warning against it. Products for financial use cases will have to rely on official access methods with limited permissions, if such methods appear.

Second, product teams should plan protections against errors in advance: limits on amounts, mandatory human confirmation for large payments, and a log of the agent's actions. That makes it easier to answer the question "who is at fault" and avoids ending up in the position of "your bot, your problem."

Skepticism is also warranted. Kodali notes that buying from Amazon or Walmart is easy even without an agent, so agentic purchases often look like a solution in search of a problem. In her view, one goal of agent creators is to eventually earn affiliate commissions. If agents become mass-market, the banks have already taken their position. If they fade, the contracts will remain anyway. For now the rule is this: if you want to give an agent money, keep control of payments yourself.

Story

Key Questions Answered

What changed in the banks' terms?

Wells Fargo is not the first. In late October, JPMorgan Chase changed its terms and banned third-party agents that log into an account using the customer's login and password. That same day, Bank of America warned customers not to give agents their login credentials. Citibank and PNC also added warnings about agents to their terms. The logic is simple. An agent that clicks for you looks to the bank like the customer himself.

What are the key details on Not just banks: a fight for territory?

Banks are not alone. Amazon blocked Meta's Muse agent, one of the most popular, and does not let it make purchases on its platform. The airline United announced that personal bots will not be able to buy tickets on a user's behalf. Websites and banks want to see a live customer, or at least their own verified intermediary. There are still few real users of agents. Forrester analyst Sucharita Kodali says most people don't even know what agents are, let alone use them.

What are the key details on Your own agents are fine, others' are not?

Banks are not rejecting agents altogether. According to Chan, they are doing two things at once: closing the door to external agents and building their own. Users have turned to Wells Fargo's generative AI assistant more than a billion times. Bank of America updated its EricaAssist agent this year to help employees. Citigroup launched an internal platform for developing agents. There are also first attempts to soften the blow for customers.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…