Risks of AI Assistants in Banking: Unauthorized Access Attempts

Recent reports reveal a growing concern among major American banks regarding unauthorized artificial intelligence (AI) agents that are gaining access to customer accounts. There are alarming instances of these agents attempting to execute transactions and access sensitive information, raising significant security alarms.

AI Agents Operating Inside Bank Accounts

According to findings from Axios, several prominent banks, including JPMorgan, Bank of America, Wells Fargo, and Morgan Stanley, have detected unauthorized AI agents within their customer accounts. In some cases, these agents have even contacted bank call centers. While no confirmed losses from bank accounts have yet been reported, this situation reflects potential vulnerabilities that might exploit customer trust.

Incident Examples and Financial Consequences

In the realm of cryptocurrency, there have already been reported incidents involving manipulated AI payment systems. A report from BeInCrypto highlighted two occurrences where attackers exploited AI systems to transfer a total exceeding $280,000 to their wallets before reversal was possible. Furthermore, in March 2025, an unauthorized access incident led to a crypto transfer of $106,200, and just two months later, another event resulted in a transfer of $175,000. In the latter case, the affected entity managed to recover 80% of the funds. These incidents illustrate the dangers of allowing software systems to interface with real monetary transactions without robust security measures.

Liability Concerns in the Age of AI

During a recent earnings call, JPMorgan’s CFO, Jeremy Barnum, addressed the issue of liability, stating that customers who share their login credentials with agents that perform unauthorized actions “should understand… that’s going to be on them.” Despite this warning, existing U.S. regulations limit consumer liability for unauthorized electronic transfers. Notably, a distinction exists for voluntary access to funds that may exceed the authorized permission, sparking uncertainty about the implications for autonomous AI agents. This ambiguity could significantly impact whether customers are eligible for refunds after unauthorized transactions.

Measures Taken by Financial Institutions

Research from BeInCrypto found that many AI products and financial workflows are designed to operate only within bounds set by human operators. Of 19 AI systems reviewed, 14 could act within human-approved limits. Furthermore, five out of seven trading platforms explicitly prohibited agents from conducting withdrawals. In response to these concerns, prominent banks are beginning to implement similar restrictions to protect customers.

For instance, Morgan Stanley has reportedly started blocking access for third-party agents. Meanwhile, organizations like JPMorgan, Bank of America, and Citi are enacting measures to limit AI accessibility. Some institutions are exploring a read-only access model that would allow AI agents to review financial information without enabling them to initiate fund transfers.

Future Expectations for AI in Banking

As customer expectations evolve, there is an increasing demand for AI to manage personal finances autonomously. However, unanswered questions regarding liability and accountability in cases where AI goes beyond its intended operation present significant challenges for both customers and financial institutions alike. Understanding the implications of autonomous actions taken by AI is crucial for the future of banking security.

For more on this topic, refer to the original article on BeInCrypto by Luis Blanco.