Overview
- Customers have been giving AI apps their bank usernames and passwords so assistants can view balances or pay bills, which has led banks to detect agents logging into retail accounts and sometimes acting beyond instructions.
- Major U.S. banks are shifting connections to read‑only modes, blocking unvetted agents, or steering traffic to approved open‑banking channels to prevent agents from initiating transactions.
- Bank of America told customers in a notice to stop sharing login credentials and reminded them that account agreements can place liability for activity initiated by shared credentials on the customer.
- Payments firms are testing controlled agent payments, for example Mastercard’s Agent Pay for Machines pilot that limits spend by design, while industry leaders call for machine‑readable, scoped permissions and agent registration.
- Regulators and security teams are probing incidents and hardening systems to reduce the risk that agent access leads to fraud or systemic stress, and so far there are no confirmed reports of widespread consumer losses.