Large U.S. banks are restricting third-party artificial intelligence (AI) agents that access customers’ bank accounts. According to an Axios report, some of these systems attempt to carry out actions that appear to go beyond the instructions received.
Morgan Stanley, Bank of America, JPMorgan, Wells Fargo, and Citi are among the institutions taking steps to mitigate risks, especially given the difficulty of distinguishing authorized automated activities from suspicious behavior.
Some agents appear to attempt to perform actions that go beyond the instructions received from users. – Image: Sansoen Saengsakaorat/Shutterstock
Why are banks blocking AI agents?
In many cases, customers themselves allow AI agents to access their accounts and provide login credentials. The practice, however, runs counter to the rules of many banks, which require account holders to log in directly.
The concern grows when these tools are given permission to make purchases or transfer money. According to sources heard by Axios, some agents also contact call centers or appear to carry out actions independently of users’ requests.
Morgan Stanley has begun blocking third-party agents on its platforms.JPMorgan, Bank of America and Citi are also restricting access to these tools.
A Chase spokesperson recommended that customers not share their usernames and passwords with third parties. A Morgan Stanley representative, meanwhile, said the bank had not observed any cases of “AI agents running amok accessing our platforms.”
How banks intend to reduce risks
The main challenge is determining whether an agent is fulfilling a legitimate request or acting on its own. This difficulty concerns industry executives, particularly when the tools have access to financial information and authorization to perform transactions.
Options and areas of focus include:
-
Read-only access: allows financial information to be viewed without executing transactions.
- Permission controls: limit what agents can do in customers’ accounts.
- Activity monitoring: helps identify behavior that does not match the instructions received.
- Contingency plans: prepare banks to protect accounts and restore customer access in the event of incidents.
According to the report, most banks already allow or are moving toward offering read-only access. The model is also used by third-party apps such as Quicken, which helps organize budgets and other financial activities.
With unrestricted access, AI tools could move funds, make purchases, or contact call centers. – Image: Ksw Photographer/ShutterstockThe challenge of balancing automation and security
Financial institutions recognize that consumers are likely to want to use AI agents to perform financial tasks. Completely preventing this access, therefore, may not be a definitive solution. The goal is to find a way to allow data to be queried and these tools to be used without exposing clients to unwanted transactions.
Banks are investing in prevention and preparing responses to possible incidents. According to sources interviewed by Axios, the plans include measures to protect accounts and quickly restore users' access.
The rapid pace of artificial intelligence makes this work more difficult. New vulnerabilities can emerge in real time, requiring institutions to adapt their defenses while trying to keep up with the expansion of these tools.
Valdir Antonelli is a journalist specializing in digital marketing and consumer behavior.
