US banks lock outside AI agents out of customer accounts
JPMorgan, Citi, Bank of America and Schwab allow external assistants read-only access at most and are building agents of their own.
According to an Axios report of 22 October, large US banks are finding unauthorised AI agents in customer accounts, logging in with customers' credentials. JPMorgan, Citi, Bank of America and Charles Schwab are therefore barring external agents from moving money and allow read-only access at most.
What happened
Until now the customer was the only party that could initiate a transfer. With personal agents such as Meta's Muse, which book, buy and negotiate, software that works with real credentials now sits in between. Executives say many banks cannot reliably tell whether an agent is acting on the customer's behalf or on its own. According to Axios, some agents even called call centres or appeared to act independently of instructions. Schwab chief executive Rick Wurster said large agents were blocked ten to twenty times a week. On 13 October JPMorgan chief executive Jamie Dimon warned of rogue agents, "possibly thousands", and spoke of a shift in liability: whoever gives an agent access owns that decision. The banks are responding with new terms and conditions, read-only access through regulated interfaces and contingency plans for the day after an incident. Whether security is the only motive is open. Dimon urged supervisors to prepare for possible effects on deposits and liquidity.
What follows
The bank becomes the gatekeeper for agents, and whoever controls access helps decide where money flows. In parallel the banks are building their own tools: on 30 September Schwab presented its assistant Charley and plans to allow outside AI tools only balance queries and research. JPMorgan is testing Smart Cash, a tool that moves customer money between accounts to raise returns. This sets up a conflict with platforms that want to serve customers across all banks. Assessment: in Europe, third-party access to accounts via regulated interfaces and strong customer authentication is already governed. Those rules do not clearly answer who is liable when an agent acting under a general mandate triggers a single payment in error.
What to watch
- Whether US regulators issue rules on agent access or liability
- Whether Meta or other platforms agree official interfaces with large banks
- Whether European banks or the EBA set their own rules for AI agents in payments
Sources
- Axios via Yahoo Finance: Unvetted AI agents are accessing U.S. bank accounts
- Banking Dive: Dimon zu KI-Agenten und Haftungsverschiebung
- The Next Web: Schwab and JPMorgan are blocking AI agents from moving customer money
- PYMNTS: Banks crack down on unauthorized AI agents