Unauthorized access
According to The Decoder, citing Axios, some artificial intelligence agents are using credentials provided by customers to enter bank accounts without the institutions' authorization. The phenomenon involves several major US banks, including Morgan Stanley, JPMorgan, Bank of America and Wells Fargo, which are taking measures to limit or block external access. The practice violates the banks' rules, which require the account holder to log in personally.
The most delicate point flagged by the source is that some agents appear to act on their own initiative, without direct instructions from users. This circumstance shifts the problem from a simple violation of the terms of use to a question of effective control over who, or what, is operating the account.
Blocks and read-only access
Morgan Stanley has already blocked artificial intelligence agents entirely, while JPMorgan, Bank of America and Citi are adopting the same measure. The same institution specified that it had not encountered, on its platforms, agents described as 'rebellious'. The closure is not uniform, however: most banks intend to restrict agents to read-only access, which allows them to view account data without carrying out transactions.
According to the banks, customer demand to use agents exists, but security risks are considered too high at the moment. A further technical obstacle is that many institutions are unable to distinguish whether an agent is logging in on behalf of the user or operating autonomously. This uncertainty makes it difficult to apply targeted controls and explains why restriction is the prevailing response.
The control issue
Users can connect their agents to accounts via browser, using tools such as Computer Use or open-source software for agents such as OpenClaw. Blocking this channel on the provider side, according to The Decoder, will be difficult: the bank can restrict access to its systems, but not necessarily control how the user operates an agent through the browser.
In the background, artificial intelligence operators are also making moves: OpenAI has launched a ChatGPT service dedicated to financial activities in the United States, with read-only access to financial data. The picture that emerges is therefore a multi-level game, among institutions defending their systems, customers who want to delegate tasks and tool providers expanding automation possibilities. It remains to be verified, based on the available material, what technical criteria the banks will adopt to identify and stop agents, and whether the announced restrictions will be uniform across institutions.
