US Banks Move to Block Third-Party AI Agents

Financial institutions are updating terms to exclude third-party AI bots, citing security risks and shifting liability to users.
Major financial institutions are setting new terms for customers using AI-powered automation, formally distancing themselves from the actions of third-party “AI agents” and shifting liability for any errors directly onto users.
Wells Fargo is the latest to update its customer agreements, with new terms set to take effect on Tuesday explicitly stating that the use of any AI agent is done “at your own risk.” This move follows a wave of policy updates across the banking sector as institutions seek to mitigate risks associated with automated software that logs into user accounts to perform tasks.
In late October, JPMorgan Chase modified its terms to prohibit third-party agents from logging in with customer credentials, while Bank of America simultaneously warned clients against sharing login information with such services. Other major institutions, including Citibank and PNC, have also updated their terms to incorporate warnings regarding the use of these tools.
The technology, while still in early stages, has gained traction among some users who utilize these bots to automate tasks such as shopping, comparing prices, or managing household orders. Although the segment of online transactions completed by these autonomous tools remains minuscule, banks are expressing concern about the security and operational risks, ranging from the potential for unauthorized transactions to errors stemming from the automated nature of these interactions.
“The banks see risks with everything,” said Sucharita Kodali, an analyst at Forrester who focuses on consumer behavior. “From the security of transactions to automated withdrawals that could happen to the detriment of account holders.”
Industry experts highlight the potential for automated mistakes that leave consumers in a difficult position. Kacy Kim, an associate professor of marketing at Bryant University, noted the inherent tension in the technology’s design: “Consumers authorize goals, not every click.” While Kim suggests these policy changes may not deter dedicated early adopters, she notes that consumers may find it unfair to be held accountable for errors made by the software they authorized.
Luca Cian, a professor at the University of Virginia who researches AI and consumer psychology, illustrated the practical risks with a hypothetical scenario involving an AI agent accidentally placing a large, incorrect order. While retailers might argue that a valid order was placed, the bank is making it clear that they do not intend to bear the responsibility for those mistakes.
Despite the harsh stance against external tools, banks are simultaneously investing in their own internal generative AI systems. Banks are “locking the door to outside agents and building their own,” Cian observed. Wells Fargo has reported over a billion interactions with its generative AI tool, Bank of America has revamped its “EricaAssist” agent for employees, and Citigroup has launched an internal development platform focused on agentic technology.
While the industry closes ranks against third-party access, some relief may be on the horizon from other quarters. American Express has signaled that it is exploring limited protections for customers, indicating that certain errors caused by agents might be eligible for reimbursement through the card provider if recovery from the merchant is not possible.
Despite the uncertainty over whether AI-driven shopping will become mainstream or fade, financial institutions have adopted a firm stance on the responsibility for these tools. Whether consumers continue to use third-party bots or not, banks have made it clear: it is your bot, and it is your problem.