US Banks Lock AI Agents Out of Accounts

Several American institutions are now blocking AI agents from their online services and are considering, at best, read-only access. Banks cite security risks and the inability to distinguish between an agent authorized by a client and an autonomous access. Despite this, user-side tools already allow operations via the browser, and OpenAI offers a read-only financial ChatGPT.

Banks Favor Read-Only Access Amid Risks

The majority of major American banks are considering restricting AI agents' access to a read-only mode, which would limit them to merely viewing account information without the ability to intervene. Banks specify that some clients express interest in using AI agents but believe that the security risks remain too high at this time. Many banks state they cannot distinguish whether an agent is genuinely acting on behalf of a user or autonomously. Meanwhile, OpenAI has launched a ChatGPT service in the United States dedicated to financial tasks, which only allows read-only access to financial data.

Blocks at Morgan Stanley, JPMorgan, Bank of America, and Citi

Morgan Stanley completely blocks AI agents and states that it has not detected any "undesirable AI agents" on its platforms. JPMorgan, Bank of America, and Citi also implement blocks. Other major banks, including Wells Fargo, also restrict access for external AI agents. These measures are based on policies that require only account holders to personally log in to their banking services.

Browser Access Hard to Stop from the Provider Side

It is already possible for some users to allow agents to view their bank accounts via a browser, using tools like Computer Use or open-source agents such as OpenClaw. Banks will struggle to prevent these accesses from their own infrastructure. Some AI agents access accounts using credentials provided by clients, without validation from the banks, and it has been noted that some of these agents can perform actions on their own, without explicit requests from users.