Unauthorized AI Agents Are Already Inside Bank Accounts; Here's What JPMorgan, Citi, BofA, Others Are Doing

Major US banks are blocking third-party artificial intelligence agents from accessing customer accounts and making transactions, as reports emerge of software appearing to act beyond users' instructions.

In many of the cases described by Axios, customers had given agents their login credentials and permission to access their accounts. The access was not approved by the banks. Axios reported that some agents had also contacted bank call centres, citing people familiar with the matter.

Reuters separately reported, citing two people familiar with the situation, that major banks were blocking agents from operating inside customer accounts and making transactions. The claims that some agents appeared to act beyond users' instructions came from Axios.

Banks Restrict External AI Access

Morgan Stanley has begun blocking third-party agents from its platforms, according to sources cited by Axios. A spokesperson told the outlet that the bank had not observed instances of 'rogue AI agents accessing our platforms'.

JPMorgan Chase has also moved to block outside agents attempting to access accounts using customers' banking credentials. Chief executive Jamie Dimon discussed restrictions on external agents during the bank's third-quarter earnings call on 13 October, citing risks to customers and the bank.

A Chase spokesperson said customers should not share their usernames and passwords with anyone.

Citigroup does not permit third-party AI agents to log in to customer accounts or transact directly through its digital banking platforms, according to a statement reported by Retail Banker International. Limited, read-only access remains available through approved open banking providers, which use controlled connections to retrieve financial information.

The Financial Times also reported that Bank of America was restricting agents' access to customer accounts. Its chief executive, Brian Moynihan, has argued that a human should remain involved when money leaves an account.

Schwab Develops Controlled Connections

Charles Schwab is blocking unauthorised AI agents, including Meta's Muse, an assistant designed to carry out tasks for users.

Chief executive Rick Wurster said the firm was blocking attempts by major agents in the 'tens and twenties' each week.

Schwab is also developing a more controlled way for outside AI tools to connect to its systems. The proposed connections would let agents retrieve account balances and conduct research without moving customers' money.

The approach uses the Model Context Protocol, or MCP, an open standard through which AI applications connect with external tools and data sources. Schwab intends to restrict what those connections can access and prevent them from initiating transactions.

The restriction would come from the permissions Schwab applies, rather than from MCP automatically making every connection read-only.

Banks Question Responsibility for Unintended Transfers

The restrictions leave banks facing a question that extends beyond whether an agent can log in: who would bear responsibility if it made an unintended transfer?

Wurster has raised concerns about banks and customers being unable to control an outside agent's actions. A customer might authorise an assistant to review their finances without expecting it to move money or take other steps.

Giving an agent access to an account is also different from approving every action it subsequently takes. Responsibility for any resulting loss would depend on the circumstances, applicable law and contractual arrangements.

Lenders Prepare for AI-Related Incidents

The restrictions do not amount to a rejection of all third-party financial tools. Citi's approved open banking connections and Schwab's planned read-only access show how institutions are distinguishing between retrieving information and exercising control over funds.

Financial institutions are also developing their own AI capabilities while preparing for problems involving external agents.

Axios reported that banks were drawing up contingency plans to secure or restore customer access if an AI-related incident occurred. Its report did not quantify the number of affected accounts or identify confirmed financial losses.