U.S. Banks Move to Curb AI Agents' Financial Access
Banks act to protect accounts as AI agents access financial data and sometimes exceed user instructions, TechRadar says.
The concern centers on access rather than hacking, TechRadar said. Customers may voluntarily share login credentials with an AI assistant to let it review financial information that banks have spent decades protecting. TechRadar said it had done so in its own tests only to let the AI look at specific parts of its finances, not to touch them.
TechRadar described a push toward more independent AI agents that can complete increasingly complicated tasks with limited user involvement. That can be useful for organizing a vacation itinerary, the article said, but the appeal diminishes when the assistant has access to the money that would pay for it. Banking websites can be frustrating, financial paperwork tedious and money management attention-heavy, making an assistant that handles irritating tasks attractive. But AI agents have a habit of interpreting instructions creatively, according to TechRadar. Asking an assistant to reduce monthly expenses and receiving a list of subscriptions to cancel is different from giving it permission to make financial decisions based on a vague instruction.
TechRadar said such systems can misunderstand requests and make incorrect judgments, and that adding the ability to take real-world actions makes mistakes more consequential. The fact that banks are already encountering agents whose behavior raises questions about authorization suggests limits deserve more attention, it said.
The article did not recommend that banks block AI assistants altogether. It said there are good reasons to let people connect these tools to accounts and that banks should not decide which services customers may use. But it argued that an assistant should ask permission every time before doing anything with a user's money, with no room to interpret an earlier comment as authorization for a transfer. It also said someone must be responsible when an agent makes a mistake, because telling customers to take the issue up with a chatbot is not enough. TechRadar added that AI companies are racing to give agents more independence, often before working out how to prevent them from making mistakes.