JPMorgan Chase, the largest bank in the United States, will not let outside AI agents sign in to retail customers’ accounts and act for them. Chief executive Jamie Dimon set out the position on the bank’s third-quarter earnings call on Tuesday, October 13, weeks after Meta launched its Muse assistant and as rival agents such as Instinct court the same users.
Dimon told analysts the bank did not want rogue agents operating on customer accounts, and warned that there could be thousands of them. A bank spokesperson then drew the line more precisely: Chase does not allow third parties to collect customers’ online banking usernames and passwords and sign in on their behalf, but it does support data sharing that customers authorize through open-banking providers.
That distinction matters more than the headline. If you have linked a bank account to an AI assistant, what still works depends on how the link was built.
Two ways an agent gets into a bank account
There are two broad routes, and they carry very different risks.
Logging in as you. The older method, often called screen scraping, has an app or agent collect your username and password and sign in to online banking exactly as you would, reading the pages it sees. Data aggregators built much of the US fintech industry this way. Plaid, whose network links apps such as Venmo and Chime to bank accounts, has historically asked users of some banks to share their login credentials for instant linking. The method works with almost any bank, but the bank cannot tell your agent from you, and whoever holds the password can, in principle, do anything you can do.
A permissioned data pipe. The newer route is an application programming interface, or API: a separate door the bank builds for approved third parties. You approve access on the bank’s own site, the agent receives a token rather than your password, and the bank decides what it can see and whether it may move money. For some banks Plaid already works this way, with users authorizing access through the bank’s portal instead of handing over credentials.
Chase is closing the first door and keeping the second open. By the bank’s account, outside agents can still read some information, such as balances, through authorized connectors, but they cannot transfer money or pay bills, and the spokesperson gave no date for when agents might be allowed to act directly.
What it means for Muse users
Muse, Meta’s personal assistant launched on September 8, connects to more than 12,000 US financial institutions through Plaid, and Meta describes that connection as read-only: the agent can report balances, transactions and investment holdings, and any action needs the user’s approval. Muse is limited to US users aged 18 and over and comes in a free tier plus plans at $20 and $100 a month.
So for a Chase customer, a read-only Muse link that runs through an authorized connection is the kind of sharing the bank says it supports. An agent that expects to log in with your Chase password and pay a bill for you is what the bank now blocks. Dimon’s warning was aimed at people who hand over credentials anyway: anyone using these agents, he said, should understand the risk and that responsibility shifts to them, which he called a liability transfer.
In plain terms, he was suggesting that if an agent you authorized with your own password does something you did not intend, the bank may not see the loss as its problem.
The US open-banking rule behind the fight
The argument sits inside an unfinished federal rule. Section 1033 of the Consumer Financial Protection Act of 2010 says a bank must make a customer’s transaction data available to that customer on request, under rules written by the Consumer Financial Protection Bureau. The CFPB issued a final rule on October 22, 2024, requiring banks to share covered data electronically with consumers and with “authorized third parties” that meet set conditions. On August 22, 2025, it reopened the question with an advance notice of proposed rulemaking covering who counts as a consumer’s “representative,” whether banks can charge fees for data access, and the security and privacy risks involved.
Those four questions map neatly onto AI agents. Is an assistant that acts for you a “representative”? Can a bank charge it to connect? Elsewhere the model is further along: Europe’s PSD2 directive in 2015 and the UK’s 2016 order to its nine largest banks forced open-banking APIs into law. The aggregator model also has a legal history. In 2021, Plaid settled class actions for $58 million over claims that it used consumers’ banking logins to gather and share data without consent, according to the company’s encyclopedia entry.
Chase is building its own agent
The bank is not against agents as such. Chief financial officer Jeremy Barnum said Chase has started a targeted pilot of Smart Cash, its own AI tool that forecasts a customer’s cash flow and shifts money between Chase accounts to improve yield. Critics will note the timing: the bank is restricting outside agents that could move deposits elsewhere while testing one that keeps them in-house.
Dimon said the bank is developing its own tools to let agents in safely, without giving a date. The next formal step on the federal side is whatever the CFPB publishes after its August 2025 notice; its rule page was last updated on August 27, 2025.
Further reading
- CFPB: Required Rulemaking on Personal Financial Data Rights
- Wikipedia: Open banking
- Wikipedia: Plaid Inc.
Image: Mysterymanblue, CC0 via Wikimedia Commons
This article was produced with AI assistance and reviewed by our Editorial Desk. How we work.
