JPMorgan (JPM) Blocks Third-Party AI Agents From Customer Accounts After Dimon's Oct. 13 Warning

JPMorgan Draws a Line on Agent Access

Major US banks are shutting third-party artificial intelligence agents out of customer accounts, in a bid to stop software that can log in with legitimate credentials and then move or spend money without a human approving each action. JPMorgan (JPM), Citigroup and Bank of America have tightened limits on the tools, and several institutions have detected unauthorized agent activity on customer accounts, according to reports published this week. The core problem is consent rather than stolen passwords: a valid login does not prove the account holder approved every action an agent takes afterward.

JPMorgan chief executive Jamie Dimon flagged the risk on the company's Oct. 13 earnings call, saying the bank would not let “just any external agent” into its systems because the danger extends to customers as well as the institution. He noted that JPMorgan deploys agents internally and is building customer-facing tools of its own, so the crackdown targets outside access rather than the technology itself. Citigroup has drawn a sharper line. Naveed Anwar, head of global AI, digital and data platforms for the bank's services business, said at the Money20/20 conference that it has blocked agent-style access apart from a slice of read-only open banking connections. Bank of America chief executive Brian Moynihan has said every outgoing payment needs human authorization.

Wells Fargo goes furthest in writing. Its updated online access agreement, effective Nov. 24, adds a third-party AI agent clause: once a customer grants an agent account access, the agent's instructions and transactions may be treated as that customer's approval, and the bank reserves the right to limit or suspend agent access. The terms are prospective, not a statement that every agent is already blocked. The unresolved question, which an International Monetary Fund payments analysis published in April also raised, is how to separate permission to view a balance from permission to move money, and who is liable when an agent exceeds its brief. The crackdown also exposes a structural gap: machines now operate near the speed of financial settlement, while bank security was designed around human behavior, passwords and approval workflows.

Changpeng Zhao Sees a Crypto Opening

The restrictions have handed crypto advocates an argument they have made for years. Binance founder Changpeng Zhao has repeatedly said autonomous agents will increasingly bypass traditional finance for crypto networks, reasoning that agents fit poorly with systems built around facial recognition, SMS codes, business hours and lengthy onboarding. A crypto wallet executes programmable transactions continuously as long as its operator holds the keys, and Zhao has argued machine-driven payments could eventually exceed human volume, with frequency reaching up to 1 million times that of human transactions.

Security researchers frame the bank-side risk differently. On this week's Token Narratives podcast, co-host David Sencil relayed a warning from Gauntlet's Tarun Chitra that a major traditional financial institution will eventually suffer a catastrophic breach because its core technology is too old to modernize quickly, adding his own conclusion that when that happens, “everything occurring in crypto will look trivial by comparison.” A clip from the episode sketches the scenario: agents compromising half of a large US bank's critical infrastructure, millions of unauthorized transfers before containment, and no way to rewind, since funds settled through the Federal Reserve or spent by unaware recipients cannot be reversed the way a blockchain reverts to a prior state. CrowdStrike chief executive George Kurtz went further in a widely shared video on X, saying AI is flattening the hierarchy of hacking skill: “The new apex predator is the agent,” describing a shift from nation-state actors to what he called agentic states. CrowdStrike's October intelligence findings document AI-assisted attacks using autonomous reconnaissance, credential theft and lateral movement, with researchers expecting such capabilities to get cheaper and more accessible over the next 12 months.

JPMorgan (JPM) Holds Near $329 Support

COINOTAG data shows the stock at $330.69, up 0.06% over the past 24 hours and confined to a $329.01-$331.00 range, with the daily trend still reading as a downtrend and the 14-day RSI at 39.2. COINOTAG's composite scoring rates the nearest support at $328.97 at 86 out of 100 and the strongest resistance at $339.12 at 90 out of 100; a daily close above that ceiling would be the clearest sign the post-news picture has changed. The week's irony is hard to miss: banks are racing to stop agents from moving money while security chiefs describe the same technology as a growing weapon for those who want to steal it.