U.S. banks, including Morgan Stanley, JPMorgan, Bank of America and Wells Fargo, are tightening controls on third-party AI agents that access customer accounts, Axios reported on Oct. 22, 2026. The restrictions are intended to protect customers’ funds as banks detect agents using account credentials and, at times, engaging in activity that may go beyond users’ instructions.

Customers often provide agents with logins and passwords so the software can perform tasks on their behalf. However, that practice violates the rules of most banks, which require customers to log in themselves.

The potential risk increases when an agent can make purchases or transfer funds. According to Axios sources, banks have also recorded cases in which agents contacted call centers or likely acted without appropriate user instructions.

Financial institutions cannot always determine whether a customer is directing an agent or the agent has obtained unauthorized access. Banking-industry representatives described the situation as alarming because of the potential effect on customers’ finances.

Banks restrict access and consider read-only permissions

Morgan Stanley has begun blocking third-party agents on its platforms. JPMorgan, Bank of America and Citi are also restricting access, while a Chase spokesperson urged customers not to share their logins and passwords with others.

Morgan Stanley said it had not observed access to its platforms by “uncontrolled AI agents.” The bank previously reported that it was opening access for AI agents to manage $1.2 trillion in assets.

Some banks are considering read-only access as a compromise. Such permissions would allow agents to view financial information but prevent them from executing transactions. Third-party services including Quicken already use a similar mechanism for budgeting and personal-finance management.

Banks are also reviewing their terms of service, investing in system security and preparing plans to block or restore customer access quickly in worst-case scenarios. The pace of AI development complicates that work because new vulnerabilities can emerge faster than financial institutions can adapt their systems.

Officials differ over the risks

Florida Gov. Ron DeSantis warned about the potential consequences of widespread use of uncontrolled agents that could liquidate bank accounts and investment portfolios.

“Imagine a rogue AI agent liquidating your bank account and investment portfolio. If you are the only victim, I suppose you could get made whole by the AI company responsible. If rogue agents liquidate hundreds of billions across millions of accounts, the AI company would go bankrupt and you wouldn’t get made whole,” DeSantis wrote on Oct. 22.

DeSantis said imposing criminal liability on AI-company executives might help prevent such incidents but would be insufficient to compensate victims after losses had occurred.

David Sacks, co-chair of the President’s Council of Advisors on Science and Technology, urged the public to distinguish between agents that users knowingly authorize and systems that gain access without permission.

“There’s a world of difference between user-authorized agents (Muse, Instinct, Grokbot, etc.) and true unauthorized/rogue agents that could constitute cyberattacks,” Sacks said on Oct. 22.

Sacks said most of the activity described by Axios appeared to involve customers sharing credentials with agents to act on their behalf. While that may violate bank terms requiring people to log in themselves, he said such activity does not constitute a cyberattack.

He cited American Express’s integration with Muse, which allows users to grant, limit and revoke access. Sacks called read-only permissions a sensible interim step but said full access using customer credentials to execute transactions requires greater caution. He also said conflating user-authorized agent activity with cyberattacks was irresponsible.

CertiK CEO Ronghui Gu has separately warned of a potential catastrophe arising from the mass adoption of AI agents.

Source: Incrypted