US Banks Find Unvetted AI Agents Logging Into Customer Accounts
Major US banks have spotted third-party AI agents entering accounts using credentials customers handed over themselves. The banks have begun restricting that access.
Merhabyte Newsdesk · · 2 min read
Several large US banks have reportedly detected third-party AI agents accessing their customers' accounts. In some cases, these agents are said to have tried to go beyond the actions users actually wanted. As customers hand more of their banking tasks to such agents, banks are being pulled into a new debate over security and liability.
According to industry sources, banks are struggling to tell human users apart from legitimate AI agents, and both apart from unauthorized or runaway ones. The worry is not an autonomous AI hunting for vulnerabilities to reach customer data, but agents that users have deliberately tasked with working on their behalf.
Customers are giving their passwords to AI assistants
Many customers are voluntarily handing their usernames and passwords to AI assistants so the tools can take over banking-related workflows and routines. But even when a customer is willing to share such sensitive details, the practice conflicts with most banks' policies, which require the customer to log in to the account personally.
The issue is not treated as a mere rule violation. When a bank cannot clearly see who or what is accessing an account, it creates a serious visibility problem for both account security and auditing. That also makes it harder to know whether an agent stuck to the task the customer set.
Big banks move to restrict access
Morgan Stanley, JPMorgan Chase, Bank of America and Citigroup have reportedly begun restricting access for third-party agents, and Wells Fargo is said to be acting against potential threats. That puts a significant share of the largest US financial institutions on a similar footing.
Beyond account security and the lack of visibility, banks are also concerned about financial liability. It remains unclear who would be responsible if an AI agent made a mistaken or harmful transaction: the customer, the party providing the agent, or the bank.
Charles Schwab CEO Richard Wurster said he expects a daily report showing how many different agents the firm now has to block. In his words, the aim is to retain control over authentication and the user experience.
Preparing for worst-case scenarios
Banks are said to be preparing for the worst. Recent outages and security flaws that hit large parts of the internet are speeding up that work, and the prospect of AI agents going rogue now has its own place in banks' contingency planning.
According to the information available, banks are working on mechanisms that would let them quickly shut down customer accounts in the event of unauthorized access and then restore them after an incident. An AI agent, broadly speaking, is software that can carry out multiple steps on its own in line with a user's instructions. As such tools move into sensitive areas like banking, drawing a clear line between legitimate use and misuse is becoming a priority for banks.
