JPMorgan AI agents ban: what Chase blocks and still allows

JPMorgan Chase will not let outside AI agents log into customer accounts and act for them, CEO Jamie Dimon said on the bank's third-quarter earnings call on October 13. Read-only data sharing through approved connectors such as Plaid still works, so an agent may be able to see your balance, but it cannot get in with your password and move money.

@CNBC broke it on X that afternoon: "JPMorgan Chase blocks outside AI agents like Muse, Instinct from directly accessing accounts." The post had more than 33,000 views by the next morning, and the replies split the way you'd guess: half say the bank is protecting customers, half say it is protecting its turf.

What the ban covers

According to CNBC's reporting and the coverage that followed, a JPMorgan spokeswoman described the rule plainly. The bank does not allow third parties to get into accounts by collecting a customer's banking credentials and logging in on their behalf. That is the "screen scraping" route, where an agent signs in as you and clicks around the website.

CNBC named two agents. Muse is Meta's consumer AI agent, which became the most downloaded free iPhone app within about 10 days of launch and reads bank data through a Plaid partnership. Instinct is a text-based digital concierge from Spear Street Technology. Neither one is accused of doing anything wrong. They're examples of the agents people are already connecting to their bank accounts.

What still works

The ban is narrower than some headlines make it sound. Per the bank's statement, JPMorgan still supports customer-authorized data sharing through open-banking providers, and it says it is working on secure connections built for AI agents. In practice:

• Read-only data like balances can still flow through Plaid or a similar connector
• Your credentials go to the connector, not to the AI agent itself
• Agents cannot move money or act inside the account
• There is no date for when outside agents might get direct, sanctioned access

Why Dimon drew the line

Dimon framed it as risk and liability. Letting just any outside agent in is "too risky for clients and for JPMorgan," he said, adding: "You don't want rogue agents, and it could be thousands of them." His warning to customers was blunter. If you hand an agent your login, "there's a liability transfer." In other words, if something goes wrong, the bank considers that your decision.

He also said regulators should prepare for agents that cause harm, and that agentic AI could affect deposits and liquidity. That last part is about money as much as safety. On September 22, after Muse launched, the S&P 500 bank index fell about 3% in one session on fears that agents would sweep idle cash out of low-yield checking into whatever pays more. A Bank of America Securities analyst wrote in September that the strategic question is who controls the agent and the customer relationship.

JPMorgan's answer is to build its own. Dimon said the bank uses AI agents internally, and CFO Jeremy Barnum said a tool called Smart Cash, which forecasts cash flow and moves money between JPMorgan accounts for better yield, is in a targeted pilot.

Our read

I think both sides of the replies have a point. Handing over your password is a bad security habit whether the recipient is a person or a program. An agent that holds your login can do anything you can do, and nobody downstream can tell whether a click came from you or from a confused model. Banks have fought fintech apps over this for years. Agents make it worse because they're built to take actions on your behalf.

It's also convenient that the safest option, an agent the bank controls, is the one that keeps deposits at the bank. My guess is other large banks land on some version of "connectors yes, logins no," while agent companies push for permissioned access in place of scraping. Meta and Sierra are already pitching one such standard, the Personal Agent Protocol.

What this means for small businesses

If you're thinking about letting an AI agent near your business money, JPMorgan's rule works as a template. Connect through official integrations and APIs, and never give an agent a shared password. Start it read-only, so it can reconcile and flag things, and keep a person on anything that moves cash. Then figure out who's liable before you switch it on. Dimon just said, on an earnings call, that if you hand an agent your login, the risk is yours.

Deciding which steps an agent can take alone and which need a human is most of the work in a good automation setup. If you want help mapping it for your own books, invoicing or payments, New Face Design's free process audit is a good place to start.