The largest U.S. banks are detecting unauthorized activity by third-party AI agents that have gained access to customers’ online accounts, Axios reported. A person from the banking industry confirmed that there have been cases in which bank customers shared their login credentials and passwords with outside intermediaries.
Banks usually detect such transactions and block them. Still, if a customer handed over their own access credentials, those transactions cannot be considered unauthorized. According to a person familiar with the matter, Morgan Stanley has blocked third-party intermediaries from accessing its platforms, although no cases of unauthorized or rogue AI agents have been recorded. Earlier this month, senior bank executives discussed the risks of AI agents acting on behalf of customers. Some of them warned that the technology carries significant risks for both customers and the banking institutions themselves.
Jamie Dimon, CEO of JPMorgan Chase, said in early October that the largest U.S. bank is working on tools to manage the risks posed by AI agents performing tasks on their own. “We will not let any outside agent get inside the company, because it carries too much risk for both customers and JPMorgan,” Dimon said.
Richard Wurster, CEO of the brokerage firm Charles Schwab, also said in early October that his institution blocks unauthorized agents. “Every day I get a report in which we track how often we have had to block individual AI agents. It rarely happens, because users are unlikely to use such solutions for their financial affairs. We block them because we want to keep full control over the authentication process and the quality of service,” Wurster said.
According to representatives of leading U.S. financial institutions, customers are using their own private AI tools, not vetted by the banks, to manage their financial accounts. “In many cases, customers simply give their username or password to an AI agent that the bank has not yet had time to check. This violates the bank’s terms of service,” explained Madison Mills, an Axios reporter who covers artificial intelligence. “In the course of our investigation, we also came across evidence indicating that these agents took actions on customers’ accounts that had not been authorized in any way,” she added.
One example of such actions not approved by the user would be an AI contacting the bank to file a complaint without first consulting the account holder. “For banks, this is the real nightmare of AI agent activity,” Mills concluded. It is worth noting, however, that these agents did not act on their own initiative or break into accounts to steal funds. Usually they simply carried out users’ instructions, without having the bank’s consent to carry out transactions on the account holder’s behalf.
BK
Source: Polski.FM
