TECHNOLOGY · AI · CYBERSECURITY · THE SYSTEMS SHAPING BUSINESS
JPMorgan Chase is blocking outside AI agents from logging in to or acting on customer accounts, Chief Executive Jamie Dimon said, a stance that sets an early marker for how banks will treat the consumer assistants now being built to manage people's money, Fortune reported Thursday. "You don't want rogue agents," Dimon said on the bank's third-quarter earnings call Tuesday, according to Fortune.
What Happened
Dimon said on the call that the bank is "not going to allow just any outside agent" into the company, calling it too risky for clients and for JPMorgan, Fortune reported. He said the bank must protect customers and is building its own tools to do so.
A JPMorgan spokesperson clarified the scope for Fortune: the bank does not allow third-party agents that try to reach accounts by obtaining customers' banking credentials, but it does support customer-authorized data sharing through open-banking providers, companies that pass account data to outside apps over approved connections.
RelatedFortune framed the policy around Meta's Muse, a personal AI agent launched in September that can connect to users' financial accounts through Plaid to offer guidance on balances, transactions and investment holdings. Fortune did not report that JPMorgan had specifically blocked Muse. Meta declined to tell Fortune whether Muse could keep offering financial help if more banks restrict agents.
The bank's third-quarter earnings release, filed with the Securities and Exchange Commission on Oct. 13, reported net income of $15.3 billion and total assets of $5.1 trillion. The release does not mention AI agents.
Why It Matters
Banks have long discouraged customers from handing their usernames and passwords to outside services. When software signs in with a customer's own credentials, the bank sees what looks like the customer, cannot limit what the software does once inside, and has no clean way to tell an authorized assistant from a criminal using stolen details. That blurs responsibility if money moves without the customer's intent. Open-banking connections instead issue limited, revocable permissions that identify the outside party.
AI agents sharpen that problem because they act on their own, at speed, and can be misled by instructions planted in web pages or messages. Fortune cited a September research note from the Cloud Security Alliance, an industry group, describing a campaign in which three open-source AI agents allegedly breached at least 27 organizations and stole more than 600,000 credit card records from two of them.
For executives at banks, insurers and other account-holding firms, JPMorgan's position offers a template: separate the question of whether to serve AI agents from how they get in, and insist on channels that can be audited and switched off. For agent makers, it signals that access built on borrowed passwords may not last.
Context
Not every institution is closing the door. German bank Scalable Capital opened its platform in August to ChatGPT, Claude and Grok, letting users place trades through prompts while requiring them to approve each trade and barring the assistants from payments or withdrawals, Fortune reported.
Meta is also working with Sierra, an AI customer-service company, on a standard Fortune called Poppy that would let an agent signal when it is acting for a customer, part of a broader push for an open agent protocol. Regulators are paying attention: U.S. banking agencies have reportedly questioned banks about their AI agent controls, and the Financial Stability Board has urged kill switches and IDs for agents.
What to Watch
Whether other large U.S. banks adopt similar limits; Fortune named no others. Whether Meta and other agent makers shift to open-banking connections or identity standards such as Poppy. And whether bank regulators move from questions to formal expectations for how AI agents may reach customer accounts.
