According to a report by the US news website Axios on Thursday, October 22, several large banks in the United States are stepping up efforts to prevent security risks posed by artificial intelligence agents (AI Agents). Banks have found that some third-party AI agents, unauthorized by the banks, are accessing customer accounts, with some agents even behaving beyond the instructions of the account holders.
Unlike general artificial intelligence tools mainly used for answering questions and generating content, AI agents can autonomously plan and execute tasks to some extent, such as data inquiries, software tool access, and executing user-requested operations. However, when these tools gain login permissions to bank accounts, they may bring new security issues, including unauthorized fund transfers and unclear transaction responsibilities.
The report by Axios pointed out that major banks like Morgan Stanley, JPMorgan Chase, Bank of America, and Wells Fargo are working to control these risks. Morgan Stanley has already started blocking third-party AI agents; JPMorgan Chase, Bank of America, and Citigroup are also limiting access of such agents to customer accounts.
In some cases, customers provide login credentials themselves and allow AI agents to access accounts; however, even with customer consent, these operations may violate the bank’s requirement for the customer to log in personally.
Bank executives are particularly concerned that AI agents with account login permissions may transfer funds or make purchases without proper supervision. Some banks have also found that AI agents contact customer service centers or exhibit behaviors that seem to go beyond user instructions.
One major challenge facing banks is determining whether account activities come from the customer themselves, authorized AI agents by the customer, or automated systems that have exceeded authorization limits.
A spokesperson for Morgan Stanley told Axios that the bank has not found instances of “rogue AI agents accessing our platforms.” The report also mentioned that several banks are considering a compromise by allowing authorized third-party tools to view account data in read-only mode but not allowing direct transfers or transactions.
In addition to account security issues, AI agents could also potentially change the way households allocate funds, thus affecting the source of bank deposits.
According to a recent report by Financial Times, the banking industry is concerned that if AI agents can automatically compare deposit interest rates at different financial institutions and transfer funds to higher-yielding accounts, it could accelerate the outflow of low-interest deposits.
Torsten Slok, Chief Economist at Apollo Global Management, warned previously that if households widely use AI agents to increase cash returns, banks may lose a significant portion of low-cost deposits used for lending.
Slok pointed out that the average annual interest rate for a checking account in the US is around 0.1%, while some other accounts offer annual rates of 3.3% to 5%. If AI agents can automatically transfer funds to higher-yielding accounts, banks may face pressure from deposit outflows.
Interest-free deposits at banks are another important source of funding to consider. With approximately $1.6 trillion in deposit holdings, assuming banks need to pay a 3% annual interest rate on these funds, the annual interest cost could be close to $48 billion.
In the face of the challenges brought by the rapid development of AI technology, banks are not only tightening third-party agents’ account access permissions but also planning contingency measures to swiftly protect accounts and restore customer access in case of security issues in related systems.
Axios noted that while the banking industry has substantial resources and extensive cybersecurity experience, the rapid development of AI technology means that new vulnerabilities may continue to emerge, and relying solely on existing security measures may not be sufficient to address all risks.