Atlas AI Resources

← Back to AI News

AI products

JPMorgan and Schwab are blocking outside AI agents from customer accounts. Here is why.

Published October 19, 2026

Two major US financial firms said last week that they are shutting out outside AI agents (AI tools that take actions for you, not just answer questions) when those agents log in with a customer’s password. JPMorgan Chase said so on its October 13 earnings call. Charles Schwab’s CEO said the same on Bloomberg Television on October 16. The question underneath both is simple: if an AI moves your money by mistake, who pays?

What happened

On JPMorgan’s third-quarter call, CEO Jamie Dimon said “you don’t want rogue agents, and there could be thousands of them,” and that the bank is “not going to allow just any outside agents to come in inside the company.” CFO Jeremy Barnum explained the mechanics. Some agent services that move money do it through credential sharing (you hand over your username and password) and screen scraping (software logs in as you and reads or clicks through the website). He called that “not secure” and said that where the bank can block it, it does.

A JPMorgan spokeswoman told CNBC that agents can still see some data read-only, such as balances, through open-banking connectors like Plaid. They cannot send money or pay bills, and there is no timeline for changing that. CNBC named Meta’s Muse agent and a texting-based concierge called Instinct as examples of the tools involved.

Schwab CEO Rick Wurster told Bloomberg: “We can’t control it, the client can’t control it, something could go wrong and who’s liable?” He said Schwab blocks the big agents “in the tens and twenties” of times per week. That is his figure, not an independent count. Schwab plans to offer MCP servers instead. MCP, or Model Context Protocol, is a standard way to plug an AI tool into another company’s system under that company’s rules. Wurster said outside AI tools will be able to check balances and do research that way, but not move money. He also pointed clients to Charley, Schwab’s own AI assistant, announced September 30.

Why it matters to a normal person

The fine print is the real story. Chase’s online banking agreement (last updated October 2026) says that if you give an agent your login, the bank “will rely on any instructions or actions taken by your Agent as if they were taken by you personally.” You are responsible for what it does, even if it “performs unintended actions or exceeds the authority you gave it,” and the bank says it will not automatically screen agent activity. Dimon’s phrase for this was a “liability transfer.” Barnum put it more simply: if the agent does something you didn’t want, “that’s going to be on them,” meaning the customer.

The practical takeaway is plain. Don’t give an AI agent your bank password. If you want AI help with money, use a read-only connection or the bank’s own tools, and keep transfer alerts turned on. The model can suggest a transfer, and it may even sound sure of itself. But confidence is not permission. Whether money can actually move should be settled by the bank’s systems and by you, not by how certain the AI sounds.

What it does not mean

It does not mean banks are rejecting AI. Both firms are building their own tools. JPMorgan is piloting Smart Cash, which predicts cash flow and moves funds, and CNBC notes this looks partly defensive. JPMorgan has fought outside apps over account access before, including over fees charged to Plaid. So security is a real concern, but so is competition for your deposits.

It also does not mean these agents have been caught draining bank accounts. The executives described risks and block counts, not specific losses. Nor is this a new law. It is company policy, and The Next Web reports that the wider question of who is liable for AI agents is still unsettled in the US.

Finally, a bank’s own assistant is not automatically right either. Schwab’s announcement warns that Charley “may provide inaccurate or incomplete results” and tells users to check that account changes went through as intended.

Sources