The clash began as consumer AI companies expanded their ambitions beyond drafting emails and booking trips. In September, Meta launched Muse with financial features built around Plaid connections, allowing users to share account balances, transactions and investment holdings for personalised guidance. Meta’s answer to corporate unease was Poppy, a proposed protocol intended to show when an AI agent is acting with a customer’s permission.¹

By Oct. 13, JPMorgan Chase had drawn a harder boundary. Chief executive Jamie Dimon said the bank would not admit outside agents into its systems simply because a customer had handed over credentials. “You don’t want rogue agents, and there could be thousands of them,” he said, warning customers that granting access also meant accepting a “liability transfer.”²

The bank is not rejecting AI outright. JPMorgan says it supports secure, customer-authorised data sharing through open-banking providers and is building its own protective tools. Its argument is straightforward: an autonomous system operating beyond the bank’s controls can make mistakes, expose data or trigger transactions whose responsibility is far from clear.¹

Charles Schwab followed with the same essential message. CEO Rick Wurster said on Oct. 14 that the brokerage was blocking unauthorised agents logging in for clients, while exploring controlled links through Model Context Protocol servers. Those connections could permit limited access to information such as balances or research, but bar higher-risk actions like moving money.²

Wurster made the liability question explicit: “We don’t want someone just giving their password to Muse, and letting Muse come in and take action on that client’s behalf.” Neither Schwab nor the client could fully control it, he said, asking: “something could go wrong and who’s liable?”³

The wider debate is not settled. Some fear AI could accelerate deposit flight by constantly hunting better rates; Wells Fargo’s Charlie Scharf calls that scenario “completely overdone.” Meanwhile, lawmakers are considering clearer accountability for agents that cause harm. For now, the major banks’ compromise is clear: AI assistance, yes; unsupervised outside agents with the keys to customer accounts, no.²