Major US banks are reportedly restricting or blocking AI agents that access financial accounts using customer credentials. Responses by Morgan Stanley and others reflect difficulties distinguishing legitimate customer delegation from unauthorized automation, alongside unclear responsibility for incorrect transactions and data access. Some providers favor read-only access, but not every bank currently permits it. The reporting concerns changes in access policies and risk management, rather than confirmation of a particular large-scale theft. Financial-agent design needs to treat user consent, provider policies and execution permissions as separate issues.