Everyone is building AI agents that can do things for you. Pay a bill, move savings, book stuff. The banks just answered: not with our money π
JPMorgan lets an outside agent see your balance if you connect it through something like Plaid, but it can't send money or pay bills. Citi doesn't let third-party agents log in or make transactions at all. Bank of America limits agent access to deposits while it figures out who is liable when an AI approves a payment. Schwab says it blocks unauthorized agents, Meta's Muse included, "in the tens and 20s" every week.
And today Axios says Morgan Stanley and Bank of America are working on how to contain the risk. The real problem is so simple it's almost funny: when an agent moves money, how do you prove the customer actually asked for it?
At the same time they are all building their own assistants. Schwab wants clients to use its Charley instead of giving their password to Muse. Of course.
Robinhood went the other way. Agents can trade there through its MCP server, with a spending limit the customer sets and manual approvals if you want them.
This is not banks being old school. This is the first place where "the agent did it" has a real price. Whoever solves the who-asked-for-this question first will be very busy next year.
The whole who-allows-what list is on PYMNTS if you want the details.
Get an email when Amy publishes