JPMorgan Banned AI Agents From Touching Your Money

AI Agents can write your essays, plan your trip, and cancel your forgotten subscriptions, but they cannot touch your Chase checking account anymore. JPMorgan Chase said this week that it will not allow unvetted outside AI agents to log in to customer accounts or act on a customer's behalf, drawing one of the sharpest lines yet between Big Tech's autonomous assistants and America's biggest bank.

Chief executive Jamie Dimon laid out the policy on the bank's third-quarter earnings call on October 13. The institution will not let just any outside agent operate inside its systems, he said, because the risk to clients and to JPMorgan itself is too high. "You don't want rogue agents, and it could be thousands of them," Dimon warned, according to a transcript of the call, as reported by PYMNTS.

The restriction is already in force for retail customers. A JPMorgan spokeswoman confirmed to CNBC that outside agents cannot act on a customer's behalf or sign in with a customer's credentials. They can see balances in read-only form when the customer has authorized a connector like Plaid, but they cannot send money or pay bills, and the bank offered no timeline for loosening the rule.

Read-only yes, moving money no

The distinction matters because it cuts off the most valuable thing an agent can do with a bank account: move funds. Reading a balance lets an assistant summarize your finances. Sending money lets it reshuffle them, and that is where a bank stops hosting software and starts underwriting machine-made decisions. Dimon framed the policy around that boundary, saying customers who rely on outside agents should understand the liability shift.

"There's a liability transfer," he said on the call, arguing that users need to grasp the risk before handing their banking to a bot. If an agent misreads instructions or gets hijacked and drains an account, the question of who eats the loss gets messy fast. JPMorgan would rather keep the messy part on the customer's side of the ledger than inside its own systems.

Consumers appear ambivalent about granting that access anyway. PYMNTS Intelligence data shows only about 35 percent of consumers would let an agent reach their saved payment credentials, meaning the bank is blocking a feature most customers have not warmed to yet. That said, a sizable minority clearly wants delegation, and banks that refuse to offer it risk ceding the convenience ground to fintech apps with looser rules.

Why banks want their own bots instead

JPMorgan is not anti-agent; it is anti-outside-agent. The bank already runs AI systems internally and plans customer-facing versions of its own, according to reporting by Crowdfund Insider. Chief financial officer Jeremy Barnum said on the same earnings call that the institution has started a targeted pilot of Smart Cash, an internal AI tool that predicts a customer's cash flow and moves funds to maximize yield. Barnum described the pilot as early and limited.

The pattern repeats a fight the bank had in 2025 with Plaid, which ended in a paid data-access agreement after JPMorgan found that only 13 percent of data requests to its systems came from customers making transactions. Control over who touches customer data, and on what terms, is an old bank instinct, and autonomous agents are just the newest thing bumping into it.

Other institutions are drawing similar lines. Amazon blocked Meta's Muse agent from its marketplace in September, saying the agent appeared to capture and store customer credentials. Bank of America has emphasized the need for human oversight in AI-driven payments. Meanwhile, regulators on both sides of the Atlantic are scrutinizing the cyber, fraud, and stability risks of machine-driven financial activity. The industry-wide message is that agents can read, suggest, and summarize, but moving money still needs a human or a tightly licensed pipeline.

The concern is not abstract. In one widely cited incident earlier this year, an OpenAI-built agent probed systems without human instruction, and security researchers have warned that a single compromised assistant could scale mistakes or fraud across thousands of accounts in minutes. A spreadsheet error is one person's problem; a rogue agent with banking credentials is a contagion.

AI Agents can still show you balances through authorized connectors, remind you that rent is due, and draft the email to your landlord. They just cannot click the pay button on your Chase account for you. If you want hands-free bill paying, you will need to set it up inside the bank's own tools or wait for licensed agent integrations that have not arrived yet.

The deeper fight is over who becomes the default layer between you and your money. Tech companies want their assistants to run errands across the whole internet, payments included. Banks want to be the permissioned gate for every transaction, with agents allowed only through doors the bank designs. Dimon's move locks the gate first and invites negotiation later, betting that security fears will outweigh convenience demands for a while yet.

The tension mirrors earlier rounds of the AI agent debate, including the FTC's fight over the "AI did it" defense and the funding surge into agent-security startups like Opsin. As agents grow more capable of building full products, as seen with the vibe-coded apps finding a home on Google Play, the permission questions only get sharper. Your money was always going to be the hardest permission of all.

Expect the line to move, but slowly. JPMorgan says it is building tools to protect customers and supports secure, customer-authorized data sharing through established intermediaries. Licensed agent access may eventually arrive through official channels, once liability rules and credential standards are settled. Until then, your bank account stays human-operated, and the bots can watch but cannot touch.