AI in action · Tools
'You Don't Want Rogue Agents, and It Could Be Thousands of Them': JPMorgan Bars Outside AI Agents From Customer Accounts
Jamie Dimon said on JPMorgan's third-quarter earnings call in October 2026 that the bank will not let any outside AI agent, such as Meta's new Muse assistant, log in to or act on customer accounts, calling it 'just too risky for clients and for JPMorgan'. Customers who hand an agent their credentials 'should understand there's a liability transfer', he said. Third-party tools can still view some data but cannot move money, according to Cybernews.
The largest bank in the United States has drawn a line that the rest of consumer finance will have to decide whether to follow. On JPMorgan Chase's third-quarter earnings call, chief executive Jamie Dimon said the bank will not allow any outside AI agent into its systems, according to Fortune and The Street: "You don't want rogue agents, and it could be thousands of them," he said, describing the prospect as "just too risky for clients and for JPMorgan". The remarks followed Meta's launch of Muse, a personal assistant designed to carry out tasks on a user's behalf across apps and accounts.
The policy, as chief financial officer Jeremy Barnum described it, is to block third-party agents from logging in to or acting on customer accounts wherever the bank can detect them, and Barnum criticised credential sharing and screen scraping, the methods such agents often use, as insecure. Cybernews reported that outside tools can still view some account data through the bank's approved data-sharing channels, but cannot send money or pay bills. Dimon said the bank is "not against agents or for agents" and uses them internally; what it will not do is accept instructions from an agent acting for a customer.
The sentence that matters for customers is about liability. People who let an agent act for them "should understand there's a liability transfer", Dimon said: if a customer gives a model their credentials and the model empties the account, the loss is the customer's under the rules that govern authorised transactions, and the bank's protections against fraud do not apply in the same way. He added that regulators should prepare for the possibility that bots and agents cause harm, a point US bank supervisors made in their own seven questions on AI agents earlier this month.
The stance is the opposite of the one taken by the brokers. eToro, Robinhood, moomoo, ThinkMarkets and Bybit have all connected AI agents to live client accounts this year, usually inside ring-fenced sub-accounts with set budgets, and Interactive Brokers lets its Claude-based assistant stage orders that a human must approve. Those firms have decided that an agent with a limited mandate is a product; JPMorgan has decided that an agent with a customer's credentials is an intruder. Both can be right, because a broker's sub-account and a bank's current account carry different risks and different legal protections.
For anyone using an AI assistant to manage money, the practical question is who the agent is logged in as. An agent operating inside a platform's own permission system, with a budget and an audit trail, is covered by that platform's rules; an agent using your password to act as you is you, with all the consequences Dimon described. The bank's refusal is the first clear statement by a systemically important institution of where it thinks that line sits.
Sources: Fortune — 'You don't want rogue agents': Jamie Dimon bars outside AI agents like Meta's Muse from JPMorgan accounts (Oct 15, 2026); The Street — JPMorgan CEO sends blunt warning on 'rogue' AI agents (Oct 2026); Cybernews — JPMorgan: AI agents blocked from customer accounts (Oct 2026)