Banks question AI agents' role

Major American banks, including Morgan Stanley, Bank of America, JPMorgan, and Wells Fargo, are expressing skepticism about AI agents, particularly their ability to make financial decisions on behalf of users. This concern arises from the difficulty in distinguishing AI agents from human users and the potential for these agents to make decisions that could be harmful to clients.

Banks seek to block AI access

As reported by Axios, these banks are seeking to prohibit AI agents from accessing their platforms. They typically require account holders to log in themselves, and they have detected unauthorized access by AI agents that can perform transactions or initiate operations without explicit user instructions. Some bank executives have described the situation as "scary."

Chase warns against sharing credentials

Chase has advised its customers not to share their login credentials and passwords with anyone, including AI agents. In contrast, Morgan Stanley has stated that it has not yet observed "uncontrolled agents accessing [its] platforms," indicating that the risk may not be universal across all institutions.

Banks aim for a middle ground

While some restrictions are being implemented, many banks are also exploring a middle ground. They allow AI agents to read information in customer accounts, such as for financial management software like Quicken, but they typically restrict actual transactions. In Europe, the directive DSP2 separates data consultation from payment execution by authorized service providers, limiting AI assistants' ability to use users' login information freely.

Bundesbank highlights risks

The Bundesbank published a report in September 2023 outlining risks associated with autonomous payments, including undesirable purchases, payment errors, and unauthorized operations. The German central bank recommends reliable identification of agents, traceability of transactions, and mechanisms for human intervention to mitigate these risks.

Secure access solutions explored

Some solutions, such as 1Password, allow AI agents to use passwords without seeing them, offering a potential workaround for secure access. This approach could help banks and users maintain security while still allowing AI agents to perform necessary functions without exposing sensitive information.