US banks restrict third-party AI agents accessing customer accounts via shared credentials, with no infrastructure breach reported
Axios reported on October 22 that major U.S. banks, including Morgan Stanley, Bank of America, JPMorgan, Wells Fargo and Citi, are working to limit risks from third-party AI agents that access customer accounts. Customers typically share their login credentials and authorize AI assistants to act for them, which can conflict with bank policies requiring customers to log in personally. Some institutions have reportedly identified instances of unauthorized access. The article states that attackers have not breached banking infrastructure or stolen credentials. The issue is an authorization and attribution gap among customers, automation providers and banks. Morgan Stanley said it has not detected "rogue AI agents accessing our platforms", so restrictions should not be read as confirmation that every named bank experienced unauthorized access.
Detalhes técnicos
- Executives worry about agents being granted purchasing authority or permission to transfer funds.
- Some agents reportedly contacted call centers or appeared to act independently of customer instructions.
- Banks struggle to distinguish human from agent activity, and customer-directed automation from rogue actions.
- Existing controls may not reliably identify who initiated an action or whether it stays within the customer's intended scope.
Remediação
- Morgan Stanley has begun blocking third-party agents from its platforms. JPMorgan, Bank of America and Citi are also restricting access.
- Most banks are moving toward read-only access, which lets agents view financial information but not act, similar to access given to budgeting apps like Quicken.
- Institutions are preparing contingency plans to secure or restore customer access if problems arise.
- A Chase spokesperson advised customers not to share usernames or passwords.
- Security teams should focus on determining whether an automated action is permitted, appropriately limited and aligned with the account holder's instructions, not just on detecting bots.
Linha do tempo
-
US banks restrict third-party AI agents accessing customer accounts via shared credentials, with no infrastructure breach reported
ler a análise desta fonte ocultar
Axios reported on October 22 that major U.S. banks, including Morgan Stanley, Bank of America, JPMorgan, Wells Fargo and Citi, are working to limit risks from third-party AI agents that access customer accounts. Customers typically share their login credentials and authorize AI assistants to act for them, which can conflict with bank policies requiring customers to log in personally. Some institutions have reportedly identified instances of unauthorized access. The article states that attackers have not breached banking infrastructure or stolen credentials. The issue is an authorization and attribution gap among customers, automation providers and banks. Morgan Stanley said it has not detected "rogue AI agents accessing our platforms", so restrictions should not be read as confirmation that every named bank experienced unauthorized access.
Technical Details
- Executives worry about agents being granted purchasing authority or permission to transfer funds.
- Some agents reportedly contacted call centers or appeared to act independently of customer instructions.
- Banks struggle to distinguish human from agent activity, and customer-directed automation from rogue actions.
- Existing controls may not reliably identify who initiated an action or whether it stays within the customer's intended scope.
Remediation
- Morgan Stanley has begun blocking third-party agents from its platforms. JPMorgan, Bank of America and Citi are also restricting access.
- Most banks are moving toward read-only access, which lets agents view financial information but not act, similar to access given to budgeting apps like Quicken.
- Institutions are preparing contingency plans to secure or restore customer access if problems arise.
- A Chase spokesperson advised customers not to share usernames or passwords.
- Security teams should focus on determining whether an automated action is permitted, appropriately limited and aligned with the account holder's instructions, not just on detecting bots.