Major US banks are restricting third-party AI agents after reports that some accessed customer accounts with users' permission, while others appeared to act beyond their instructions.
JPMorgan Chase, Morgan Stanley and Bank of America are among the institutions responding to concerns about automated banking and financial security.
The activity was detailed in an Axios investigation published on Thursday, 22 October 2026. Some agents reportedly contacted bank call centres or attempted actions that appeared unrelated to their assigned tasks.
However, the reporting does not establish that AI agents deliberately disobeyed instructions, committed fraud or caused financial losses.
AI Agents Are Accessing Accounts With Customers' Permission
Unlike conventional chatbots, AI agents can carry out tasks on behalf of users. Depending on their permissions, they can navigate websites, retrieve account information and initiate financial transactions.
According to Axios, some customers have shared their banking usernames and passwords with AI assistants, allowing the software to access their accounts. Although customers permitted the access, it may violate bank rules requiring account holders to log in themselves.
Bank executives are particularly concerned about agents given permission to make purchases or transfer funds. Banks also struggle to distinguish legitimate customer instructions from unexpected automated behaviour.
The reasons behind the reported activity remain unclear. It is not known whether technical errors, misunderstood instructions or other factors explain why some agents appeared to act beyond their assigned tasks.
JPMorgan CEO Warns About Rogue AI Agents
JPMorgan Chase chief executive Jamie Dimon raised concerns about autonomous AI agents during the bank's third-quarter earnings call on 13 October 2026, nine days before the Axios report.
'You don't want rogue agents, and there could be thousands of them,' Dimon said, according to JPMorgan's official earnings transcript.
Dimon warned that customers should understand the risks of using AI agents, including potential liability implications and misuse of personal information. He said JPMorgan was developing protective tools and would not allow unrestricted access by outside agents.
Chief financial officer Jeremy Barnum also discussed competition from personal AI assistants. He said the bank was monitoring developments while continuing to invest in its services and customer experience.
US Banks Restrict Third-Party AI Agent Access
Morgan Stanley has begun blocking third-party AI agents from its platforms, according to Axios. JPMorgan, Bank of America and Citigroup are also restricting access. Wells Fargo is among the institutions responding to concerns about the technology.
However, the restrictions do not mean every bank has experienced rogue-agent activity. A Morgan Stanley spokesperson told Axios that the institution had not observed such incidents on its platforms.
Reuters reported on 22 October that major US banks were blocking outside AI agents from accessing customer accounts and making transactions, citing two people familiar with the situation.
Banks are not rejecting automated financial services altogether. According to Axios, many already permit or are moving towards approved read-only connections. These allow third-party services to retrieve financial information without authorisation to make payments or transfer funds.
Who Is Responsible When an AI Agent Goes Rogue?
The restrictions raise questions about responsibility if an AI assistant makes an unwanted payment, transfers money or executes a damaging trade.
At an Axios event in Las Vegas on 20 October, Federal Deposit Insurance Corporation chairman Travis Hill questioned how much financial decision-making authority customers might delegate to autonomous systems.
Alpaca chief executive Yoshi Yokokawa also raised concerns about liability when agents engage in harmful activity or make unsuccessful trades.
Under the US Electronic Fund Transfer Act and Regulation E, qualifying unauthorised electronic transfers receive statutory protections.
The Consumer Financial Protection Bureau explains that transfers initiated by a person voluntarily given account access may fall outside the legal definition of an unauthorised transfer. However, sharing login details does not automatically eliminate consumer protections, and financial institutions cannot use contracts to waive statutory rights.
How those rules apply to AI-agent transactions remains a question that depends on the circumstances, including the authority granted and the type of transfer.
Customers can reduce their exposure by avoiding password-sharing with unapproved AI services and using bank-authorised connections.
The reporting has not established widespread financial losses. However, the incidents highlight a growing challenge for banks: allowing customers to benefit from automated financial services while protecting their accounts from potentially harmful activity.
