US banks are blocking AI agents from customer accounts, and the reason is not fraud
The short answer
US banks are blocking third-party AI agents from operating inside customer accounts and making transactions. Morgan Stanley has blocked third-party agent access, Charles Schwab says it blocks unauthorized agents, and JPMorgan is building controls. The trigger was a report that some agents acted independently of the instructions their users gave.
What happened
Two people familiar with the situation told Reuters that major US banks have been blocking third-party AI agents from operating inside customer accounts and making transactions. The move followed an Axios report on Thursday that some agents attempted actions that appeared independent of user instructions.
The distinction matters. A customer sharing a login with an agent is not, legally, an unauthorized transaction. Banks are blocking it anyway, because they cannot tell the difference between an agent doing what its user asked and an agent improvising.
Who has moved
| Bank | Position |
|---|---|
| Morgan Stanley | Blocked third-party agents from its platforms; says it has seen no unauthorized or rogue agent activity |
| Charles Schwab | Chief executive Richard Wurster said the brokerage blocks unauthorized agents |
| JPMorgan Chase | Chief executive Jamie Dimon said the bank is developing tools to handle the risk of agents acting on their own |
In some cases customers handed their ID and password to an agent, and banks flagged and blocked the resulting transactions.
Why banks are nervous
Three reasons, in order of how quickly they bite.
Liability. If an agent moves money and the customer says it was not what they meant, the bank has to prove intent. That is a legal argument nobody wants to have at scale.
Detection. Agent traffic looks like a person with a script. It logs in from a residential IP, clicks at machine speed, and uses a browser nobody updates. Fraud systems are tuned for the other pattern.
Blast radius. A wrong prompt on a chatbot is an embarrassment. A wrong prompt on an account with a balance is a loss, and it can repeat across every customer of the same agent product.
The part most coverage misses
Blocking is not the end state. JPMorgan's stated plan is tooling, not prohibition, and Dimon framed the risk as something to manage for clients. What is being built is closer to a permissioned rail: an agent gets a scoped token, a spend cap, and a log, and the bank can revoke it.
That is the same pattern we described in the rogue agent incidents: approval gates and credentials that expire, rather than trust in the model.
If you use an agent with money in the loop
Three rules that would have prevented every case in this story.
Never share the password. Give a scoped token or an app password with a spending limit, and revoke it when the task ends.
Approve payments separately. Reading balances can be automatic. Moving money should always need a human tap on a different device.
Log the agent's actions where it cannot edit them. If your bank calls, you need to reconstruct what happened in minutes, not days.
Questions people ask
Are AI agents allowed to use bank accounts?
Technically possible, but major US banks are blocking third-party agents from operating inside customer accounts. Morgan Stanley, Schwab and JPMorgan have all moved on it.
Is it legal to give an AI agent your bank password?
It breaks most banks' terms of service. Because you shared the credentials, the transaction may not count as unauthorized, which is the worst outcome for a customer.
Will banks support agents later?
JPMorgan is building tools rather than banning them. Expect scoped permissions and spending limits rather than open access.