Skip to content
Rush Commerce
AI & Automation3 min read

JPMorgan blocks outside AI agents: give agents their own keys

JPMorgan is blocking outside AI agents that log in with customer credentials or screen-scrape. Who owns the liability, and how to wire agents to bank data safely.

JPMorgan Chase is blocking outside AI agents from customer accounts wherever it can. On its third-quarter earnings call on October 13, CEO Jamie Dimon and CFO Jeremy Barnum said the largest US bank will not let "just any outside agent" in, and singled out agents that log in with a customer's password or screen-scrape the site. If your bookkeeping, cash-flow or reconciliation automation works by handing a bot your bank login, this is aimed at you.

What actually happened

As reported by Business Insider, Barnum said the bank is blocking outside agents "to the extent that we're able to block those." He called out two practices specifically: credential sharing, which he said the bank has long opposed, and screen scraping. His verdict: "It's not secure."

Barnum also put the liability on the table. A customer who gives an agent their credentials could be responsible for whatever that agent does that they did not want. He called it "not an ideal state of things."

Dimon framed it as scale risk: "you don't want rogue agents, and there could be thousands of them." He described a "liability transfer" away from the bank when users run their own agents, and said JPMorgan is building its own agentic tools for clients. The bank has also tightened controls on the AI tools its own employees use.

Why outside AI agents with your credentials are the problem

Read that liability line twice. When an agent logs in as you, the bank sees you. Every transfer, every bill pay, every changed setting is legally your action. If the agent is prompt-injected by a poisoned invoice PDF or just misreads a balance, there is no "the AI did it" clause in your deposit agreement.

Screen scraping is also fragile on its own terms. A bank that is actively blocking scrapers will break your automation without notice, usually on the day you need the month-end numbers.

The fix is not "stop using agents." It is to stop letting agents borrow your identity:

Use token-based connections, not passwords. Read-only access through your bank's API or an established aggregation provider gives the agent a scoped key you can revoke, not your login.

Separate read from move. Let the agent read transactions and draft payments. Keep the "send" step behind a human approval, in the bank's own interface.

Inventory what already has your login. Old bookkeeping tools, browser extensions and "AI assistants" with saved passwords are the first things to audit and rotate.

Key takeaways

  • JPMorgan says it is blocking outside AI agents that use customer credentials or screen scraping, where it can
  • CFO Jeremy Barnum: a customer who hands an agent their credentials may own the agent's unwanted actions
  • Dimon cited the risk of "thousands" of rogue agents and said the bank is building its own client agents
  • Screen-scraping automations against a bank that blocks them will break without warning
  • Give agents scoped, revocable, read-only keys and keep money movement behind a human approval

Your agent should never be holding your bank password. We build finance automations that read through scoped API connections, draft the work, and stop at a human approval before money moves. See how we wire agents into your back office, or send us the tools that touch your accounts and we will map what has your login.

Sources: Business Insider via Yahoo Finance, JPMorgan Q3 2026 earnings call highlights.

  • #ai-agents
  • #jpmorgan
  • #banking
  • #credentials
  • #agent-security
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.