Banks move to restrict AI agents as liability for transfers remains unclear

Giving an AI assistant your banking password to check a balance or make a transfer is something several users are apparently doing.

Unauthorized AI agents are being detected by major US banks operating inside customer accounts, Axios reports, with some agents even making contact with bank call centers.

The institutions responding to the issue include JPMorgan, Bank of America, Wells Fargo, and Morgan Stanley.

Confirmed bank-account losses were not documented by Axios. Crypto, however, has already demonstrated the high costs associated with manipulated AI payments.

Two incidents where attackers caused AI-linked systems to send $281,200 to their wallets prior to recoveries have been documented.

During March 2025, an attacker breached AIXBT's dashboard to trigger a $106,200 crypto transfer. A malicious message processed through Grok caused Bankr to transfer $175,000 in tokens two months later, with Bankr subsequently reporting an 80% recovery of those funds.

Both incidents, while happening in crypto, exposed the dangers of granting software access to real money without adequate safeguards.

If an AI Moves Your Funds, Who Bears the Loss?

A warning was offered by JPMorgan CFO Jeremy Barnum during the bank's October 13 earnings call.

According to Barnum, customers who share their login credentials with agents that take unwanted actions should understand "that's going to be on them."

Consumer liability for qualifying unauthorized electronic transfers is limited by US Regulation, though liability can arise when a person who voluntarily authorized a money transfer exceeds that permission.

It remains uncertain how those rules apply to autonomous AI agents. The distinction could ultimately decide if a customer receives a refund.

Why the Largest US Banks Are Blocking AI Agents

An industry review found 14 of 19 AI products and financial workflows could operate within human-approved limits. Five of seven trading venues reviewed explicitly barred agents from making withdrawals.

Comparable restrictions are being introduced by banks.

Morgan Stanley has reportedly moved to block third-party agents, while JPMorgan, Bank of America, and Citi are restricting access. Morgan Stanley reports it hasn't observed any rogue agents on its platforms.

A read-only access model is being adopted by some banks, letting agents check financial information but not move money.

AI agents that chase better rates for idle deposits could wipe roughly $500bn of equity value off US banks.

What we are seeing now tho is potentially unauthorized actions, but could also be agents who have been given unlimited access to act on the owners behalf.

While customers may eventually anticipate AI handling their finances, the unresolved question currently is who bears the loss when software acts beyond its instructions.