Banks move against AI agents as 2026 adoption stumbles

Wells Fargo will enforce an updated agreement with millions of customers starting Tuesday, specifically about AI agents. The bank’s warning to early adopters leaves no ambiguity: “Your use of any AI Agent is AT YOUR OWN RISK,” the new terms state.

AI agents, which automate online tasks and click on a user’s behalf, have stumbled through 2026. They remain nascent technology, but a growing subset of power users adopted personal bots this year. Sometimes these early adopters enjoyed the efficient help of automation, and other times they publicly embarrassed themselves when the AI agents made major mistakes, like posting bank details for everyone at work to see.

Builders of these AI tools often market agents as the ultimate tool for shopping decisions, whether that is putting together a grocery order, finding accommodations for a vacation, or discovering specific holiday gifts for parents. Some users value these agents so much that they are oblivious to overarching security concerns, pairing account credentials and granting permissions for the agents to take sweeping actions without forethought. Although the segment of online transactions completed by these autonomous tools remains minuscule, US-based banking institutions are already tired of third-party AI agents.

It is not just banks; AI agents are igniting turf wars around the web. Amazon, one of the largest online retailers, blocked Meta’s Muse, one of the most widely used AI agents, from making purchases on its platform. Travel companies are also pushing back—United recently announced that personal bots would be blocked from purchasing flights on a user’s behalf.

As the end of the year approaches, multiple banks have enacted new policies around agents or sent warnings. In late October, JPMorgan Chase changed its terms to ban third-party agents that log in with user credentials. That same day, Bank of America warned its customers not to share login info with agents. Along with the announced change coming from Wells Fargo, Citibank and PNC have also changed their terms to include warnings about agents.

The automation, which remains a central selling point of always-on agents, is exactly what may get users in trouble. “Consumers authorize goals, not every click,” says Kacy Kim, an associate professor of marketing at Bryant University who studies consumer behavior and emerging tech. While Kim expects some consumers to feel like it is unfair to be held accountable for their agent’s mistakes, she does not expect these policy moves by financial institutions to scare off the early-adopter crowd already deep in using the tools as part of everyday life.

Banks appear to be trying to get ahead of major issues, in case agentic tools stick around and do not just disappear as another tech fad. “All of our data shows that most people don't even know what agents are, much less use them,” says Sucharita Kodali, a Forrester analyst who focuses on ecommerce and consumer behavior. She sees banks as fundamentally risk-averse, so taking proactive measures now makes sense, even if most agentic issues remain edge cases.

“The banks see risks with everything,” she says. “From the security of transactions to automated withdrawals that could happen to the detriment of account holders.”

Even though outliers exist, many users are aware of the risks and are only comfortable handing over so much control to an AI agent, and only for certain shopping tasks. “People are very comfortable with AI buying toilet paper,” says Luca Cian, a professor at the University of Virginia who studies consumer psychology and AI. He sees consumers as increasingly comfortable turning to AI tools for shopping advice but often only automating boring buys, like household supplies. Shoppers are less comfortable automating away dynamic, big-ticket purchases that require additional care and discretion.

Cian gives a hypothetical scenario where a customer’s AI agent accidentally buys 10 roasted chickens from the supermarket. Who is to blame? The retailer would likely argue that they completed a valid order that the user placed. The AI builder may claim that they just provided the software, and even included disclaimers about potential mistakes. And now the bank is making it clear they do not want to be held responsible either. Hope the user needs the protein, because it looks like 10 hot chickens were bought.

Certain credit cards are looking into limited protections for shoppers. Mainly, American Express announced that some agentic mistakes, like ordering too many of a certain item, may be eligible for reimbursement in the future through the card provider if the user cannot get money back from the original retailer.

Banks are pumping out reasonable warnings about third-party agents, but do not get it twisted. These banks are all-in on first-party tools. “Banks are doing two things at once,” Cian says. “They are locking the door to outside agents and building their own.” Users have interacted with Wells Fargo’s tool that answers questions using generative AI over a billion times. Bank of America revamped its new EricaAssist agent this year to help employees. Citigroup even launched an internal development platform specifically centered on agents.

Despite all the hype, Kodali remains skeptical about whether all of this hoopla is even necessary. “The issue with agents and shopping is that, first of all, it is not difficult to buy on Amazon or Walmart or anywhere else,” she says. “In most of those transactions, you don't need an agent.” Agentic shopping often seems to be a solution in search of a problem or at least an attempt by AI builders to find a way to earn affiliate revenue eventually from their software tools.

Whether AI agents grow in popularity for shopping automation or slowly fade into the digital graveyard, banks are already resolute on the matter. It is your bot, and your problem.

Bank policies tighten as agent mistakes make headlines

Wells Fargo's updated agreement signals a shift in how financial institutions view third-party AI agents. The “AT YOUR OWN RISK” phrasing in new terms makes clear the bank disclaims responsibility for agent actions. This follows a pattern: JPMorgan Chase, Bank of America, Citibank and PNC all introduced or updated warnings about agents in October alone.

Retailers and travel companies push back

Amazon blocked Meta’s Muse from making purchases on its platform. United announced personal bots would be blocked from buying flights. These rejections highlight that not all platforms welcome autonomous agents, even as AI builders promote them as shopping aids.

Consumers comfortable only with mundane automations

Luca Cian's research shows a split in user comfort levels. People readily accept AI buying toilet paper or household supplies. Big-ticket or dynamic purchases—like groceries that vary daily—draw more hesitation. The 10-chicken hypothetical illustrates how blame could spread among retailer, AI builder and bank, leaving the user to absorb the cost.

Banks build first-party tools while blocking outside agents

Banks are not only restricting third-party agents; they are investing in their own. Wells Fargo's generative AI tool has been interacted with over a billion times. Bank of America revamped EricaAssist to help employees. Citigroup launched an internal development platform centered on agents. The message is clear: outside agents face restrictions, in-house tools get full support.

American Express offers limited reimbursement

American Express said some agentic mistakes, like ordering too many of a certain item, may be eligible for future reimbursement through the card provider if the user cannot recover money from the original retailer. This is a narrow safety net compared to the broad risks banks are warning about.

The real problem, banks say, is yours

Forrester's Kodali notes that most people do not even know what agents are, much less use them. Banks are risk-averse and acting proactively even though many agentic issues remain edge cases. The takeaway from every warning and policy change: the bot is the user's responsibility.