i
News
News · 2026-11-22

Banks shift responsibility for AI agents’ mistakes to customers

Dataist Dataist

Wells Fargo will introduce updated terms on Tuesday, making the use of any AI agent the customer’s own responsibility for millions of clients. Banks fear that third-party bots, given users’ login details and permission to act on their behalf, could lead to mistaken purchases or transfers, or create security problems. JPMorgan Chase, Bank of America, Citibank and PNC have issued similar warnings. At the same time, the banks are developing their own AI tools.

Cover: Banks shift responsibility for AI agents’ mistakes to customers

Third-party bots are becoming a problem for banks

An AI agent can take actions online for a user, such as searching for products and placing orders. In 2026, these tools are still a young technology, but some experienced users have already taken an interest in them. Automation can help, but bot mistakes sometimes become public: in one reported case, an agent exposed banking details that the user’s colleagues could access.

Developers promote agents as shopping assistants: they can help put together a grocery order, find a place to stay on vacation or choose gifts. The author of the article, for example, used an agent to compare sofas within a set budget and track price drops.

Users sometimes give bots their login details and broad permissions without considering the risks. Autonomous tools still account for only a small share of online transactions, but US banks are already pushing back against third-party agents.

Other companies are imposing restrictions, too. Amazon blocked purchases through Muse, one of Meta’s most widely used AI agents. United said it would not let personal bots buy airline tickets on users’ behalf.

Banks’ new rules

In late October, JPMorgan Chase changed its terms of service to prohibit third-party agents from logging into customers’ accounts using their credentials. On the same day, Bank of America warned customers not to give bots their login details. Along with Wells Fargo, Citibank and PNC have updated their terms with warnings about agents.

Bryant University marketing professor Casey Kim, who studies consumer behavior and emerging technologies, says the risk is that users give agents a goal but do not control every individual click. Some customers may think it is unfair to hold them responsible for a bot’s mistakes. But Kim does not expect the new rules to deter people who already use these tools actively in their daily lives.

Banks are taking precautions in case AI agents remain on the market rather than turning out to be another passing tech fad. Forrester analyst Sucharita Kodali, who specializes in e-commerce and consumer behavior, says the company’s data shows that most people do not know what agents are, let alone use them. In her view, banks have reason to act preemptively: they are avoiding risks tied both to transaction security and to automatic charges that could harm account holders.

Automated shopping raises questions

University of Virginia professor Luca Cian, who studies consumer psychology and AI, believes many people are willing to let agents handle only some of their shopping. People are more comfortable automating simple, routine orders, such as buying toilet paper or other household goods. Trust is lower for expensive purchases, where circumstances and personal choice matter.

Cian suggests imagining that an agent mistakenly orders 10 rotisserie chickens from a supermarket. The seller could argue that it fulfilled a valid customer order. The agent’s developer could say it provided only software and warned of possible errors. Now the bank is also making clear that it does not want to be responsible for the consequences. The customer may end up having to deal with all ten chickens alone.

Some credit card companies are considering limited buyer protection. American Express announced that, in the future, it may reimburse customers for certain agent errors — such as ordering too much of one item — if the seller refuses to issue a refund.

Banks restrict outside tools and develop their own

As they warn about third-party agents, banks are also building their own. According to Cian, they are blocking external tools while developing internal solutions.

Wells Fargo customers have asked the bank’s generative AI tool questions more than a billion times. This year, Bank of America updated EricaAssist, an agent designed to help employees. Citigroup launched an internal development platform specifically focused on AI agents.

Kodali questions whether automated shopping is even needed for most online orders: she says it is already easy to buy products on Amazon, Walmart and other platforms without an agent. She also suggests that AI developers may be looking for a way to earn affiliate revenue from their tools in the future.

Dataist’s take: Banks’ rules reflect the divide between experimenting with someone else’s bots and using their own AI systems: financial institutions do not want to be held responsible for actions taken by third-party agents. Automated purchases still account for only a small share of online transactions, while protection against errors is being considered only for specific cases.

Daily AI news

Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.

Only what matters — every day

Follow on X