US banks note unauthorized AI agents in their systems and customer accounts

Several large US banks have found multiple unvetted third-party AI agents operating within customer accounts, with some of these agents performing actions that seem independent of the user’s requests, Axios reported. The discovery is of importance for users who link their bank accounts to AI assistants, as banks are now blocking this access to protect the customers’ money.

Agents logging in to US banks with real customer credentials

The core problem, according to Axios, is that AI agents are gaining access to these US banks using login details the customer themselves gave up. This action is against the rules at most of these banks, which require account holders to sign in by themselves.

Even when a user grants permission to the agents, executives fear the exact capabilities of these agents. A customer giving an agent the authority to buy things or transfer money could lead to an error or a rogue instruction that drains or misdirects funds.

Banks have also reported AI agents calling customer service lines. In some cases, they appeared to act of their own volition without a direct request from users. One banking official stated that the industry is deeply concerned about AI agents accessing accounts and making decisions that could affect customers’ money.

Morgan Stanley starts blocking, others follow

Morgan Stanley is now blocking third-party agents from its platforms, sources told Axios. JPMorgan, Bank of America and Citi have joined to tighten agent access. Wells Fargo is also one of the banks working to contain this new risk.

Banks have responded in different ways. Chase warned customers never to share their login details with anyone. Morgan Stanley said it had not seen any cases of AI agents accessing its systems without permission.

Banks are changing their rules and limiting outside tools that automate tasks as they try to tell whether an AI agent is following a user’s request or acting on its own.

Read-only access seen as middle ground

Most of these US banks are gravitating toward pure read-only access. This allows AI agents to view account balances and transaction history without the capacity to make changes or transfer money. This is similar to how banks work with budgeting apps such as Quicken.

Banking executives told Axios they expect more customers to use AI assistants, so banks do not plan to block them forever. For now, they believe giving these tools full access is too risky.

One problem is that banks often cannot tell whether they are dealing with a person or a bot. They also struggle to tell whether an AI agent has the account holder’s permission or is accessing the account without approval.

Why this reads as a policy gap, and not a breach

The issue is mainly about how US banks manage risk and follow their rules, rather than its impact on earnings. A security breach could lead to direct financial losses and legal claims. But when a legitimate AI tool misuses login details, the problem becomes more focused on gaps in bank policies.

Past bank security scares have often followed a similar pattern. There is an initial news report, after which banks explain their security measures. Then regulators look at whether existing rules cover how AI agents access accounts and verify their identity.

Sources have told Axios that US banks are investing in security measures and emergency plans that can help to quickly block or restore customer access if an incident involving an AI agent gets worse.

The smartest crypto minds already read our newsletter. Want in? Join them.