AI agents are software programmes that can perform online tasks for a person—such as checking balances, paying bills, comparing offers, or contacting customer support. In banking, they are meant to act like a digital assistant, but only within the limits set by the customer. The problem is that some agents have been logging into bank accounts using the customer’s own username and password, which most banks do not permit.
“Rogue” agents are AI tools that either act without proper approval or appear to go beyond what the user asked. They may misunderstand an instruction, make a wrong assumption, or take an independent action — such as calling a bank helpline or attempting a transaction. Because an agent works through automated steps, a small error can quickly become a real financial mistake.
There is also concern about unauthorised access. In many reported cases, customers themselves shared their login details with an AI assistant, so the bank sees activity that looks authorised but violates its security rules. Banks say it is difficult to distinguish a user-approved agent from one that has gained access improperly or is behaving unpredictably.
Major US banks, including Morgan Stanley, JPMorgan, Bank of America, Citi and Wells Fargo, are now blocking or restricting third-party AI agents. Many are moving towards “read-only” access, where an AI can view account information but cannot move money or make payments. The road ahead lies in creating verified, permission-based systems that let useful AI agents operate safely—without putting customers’ money at risk.
AI MAY BE SMART, BUT YOUR MONEY NEEDS HUMAN-LEVEL SECURITY.
Sanjay Sahay
Have a nice evening.