JPMorgan and Other Major Banks Restrict Outside AI Agents

JPMorgan Chase, Citigroup and other major U.S. financial institutions are restricting third-party artificial intelligence agents from accessing customer accounts and initiating transactions, according to reports cited by Bitcoin.com News. The measures target a security gap in which software can possess a valid password without having clear approval for every action it takes.

Reuters reported on Oct. 22 that banks were blocking outside agents from operating inside accounts and making transactions, citing two people familiar with the situation. In some cases, customers had supplied their login credentials, with banks subsequently flagging and blocking transactions, a banking industry source told Reuters.

The distinction is between authenticating an account login and authorizing a particular financial decision. Giving an assistant access to banking information does not necessarily establish that its owner intended to approve a transfer, purchase or other action the software might attempt.

Citigroup has said it does not permit outside AI agents or assistants to log into customer accounts or transact directly on its digital banking platforms, USA TODAY reported. Certain Citi-approved open-banking providers can nevertheless receive limited, read-only access, preserving a channel for financial information without granting transaction authority.

JPMorgan chief executive Jamie Dimon raised concerns about unrestricted outside agents during the bank’s Oct. 13 earnings call, Bitcoin.com News reported. Bank of America chief executive Brian Moynihan has also emphasized human authorization for outgoing payments, placing customer approval at the center of the industry’s response.

Wells Fargo told USA TODAY it had taken steps to block unauthorized agents from customer accounts. Its updated access agreement, scheduled for Nov. 24, distinguishes between viewing financial information and carrying out transactions, according to Bitcoin.com News.

The restrictions do not amount to a rejection of AI across banking. Citi’s 2025 annual report describes plans for broader integration of generative AI, including autonomous agents, while warning that inadequate management could increase risks and costs. The same report identifies unauthorized transactions, exposure of sensitive information and misappropriation of funds as potential consequences of increasingly sophisticated AI misuse.

That creates a challenge for banks seeking both automation and tighter controls. An assistant that helps a customer understand spending presents a different permission question from one that can withdraw or redirect funds. Read-only connections offer a narrower compromise, although they do not resolve how banks should handle more ambitious customer requests.

Bitcoin.com News also highlighted arguments from Binance founder Changpeng Zhao that cryptocurrency networks could become attractive payment infrastructure for autonomous agents. That remains an outlook rather than an established outcome, and programmable payments do not by themselves demonstrate that an agent has correctly understood its owner’s wishes.

The next test is whether banks and software providers can establish access arrangements that clearly separate information gathering from financial execution. For now, the reported policies favor approved connections and explicit human control over payments, rather than treating possession of a customer’s credentials as unrestricted authority to act.