JPMorgan blocks AI agents from logging in and paying — but data sharing remains
The bank is shutting out AI agents like Meta's Muse from logging into or transacting on customer accounts — but customer-controlled data sharing through open banking remains. CEO Jamie Dimon points to "rogue agents" and the risk that customers end up footing the bill.
What happened
At JPMorgan Chase's third-quarter earnings presentation on October 13, CEO Jamie Dimon said the bank is blocking external AI agents from logging into or transacting on customer accounts. Fortune, via Yahoo Finance, reports Dimon's remarks from the presentation: "We're not going to allow just any outside agent to come inside the company, because that's just too risky for clients and for JPMorgan. You don't want rogue agents" (Fortune via Yahoo Finance).
In practice, this means services such as Meta's personal assistant Muse and the agent service Instinct can no longer gain direct access to retail customers' Chase accounts and act on their behalf, according to CNBC, as reported by Mashable (Mashable). A Chase spokesperson confirmed to CNBC that third parties cannot obtain customers' banking credentials and log in on their behalf. The bank has given no timeline for when, or whether, direct agent interaction will be permitted.
It is worth noting that all coverage of Dimon's remarks rests on secondary reporting of the presentation — no bank document or complete transcript has been made available. The remarks were reported by Fortune, CNBC via Mashable, and 24/7 Wall St., and are mutually consistent.
What the agents can still do — and what they can't
The block is not absolute. The bank still supports secure, customer-approved data sharing through open banking and approved connections. An agent can therefore still retrieve information such as account balances, as long as the customer has consented. But this access does not extend to sending money or paying bills, the Chase spokesperson told CNBC, according to Mashable (Mashable).
The line-drawing is no accident. When Meta launched Muse on September 8, the agent was described as a personal assistant that can book travel, fill out forms and shop online. Through an integration with the fintech company Plaid, Muse users can connect to financial accounts and receive personalized guidance on balances, transactions and investment holdings, according to a Plaid press release as reported by Fortune (Fortune via Yahoo Finance). It is precisely this type of read-only, consent-based data access that still gets through: the agent can read and advise, but not log in as the customer or move money.
Why the bank draws the line where it does
According to the secondary reporting, Dimon pointed to two factors. The first is security: the risk of letting uncontrolled agents into a banking system. The second is liability. According to 24/7 Wall St., he highlighted concern over a possible "liability transfer" — that customers could end up bearing the costs of an erroneous payment an agent has made (24/7 Wall St.). The question of who covers losses when an autonomous service pays the wrong account, at the wrong time or in the wrong amount is unresolved in general; Dimon's statement is a concern, not a settled regulatory framework.
The context is not theoretical. Over the autumn, the disagreement between agent developers and platforms has already become visible. Amazon blocked Muse roughly two weeks after its September 8 launch. Amazon claimed that Meta did not notify the company of the integration, and that Muse does not identify itself as an AI agent while navigating the website, according to 24/7 Wall St. — that is, an agent behaving like a human user, without a declared identity.
The security community has also begun pointing to agents as an attack tool. A note from the Cloud Security Alliance, reported by Fortune, summarizes an alleged cybercrime campaign in which three open-source AI agents purportedly collaborated to compromise at least 27 organizations and steal over 600,000 credit card records from two of them (Fortune via Yahoo Finance). This has not, however, been independently confirmed, and should be read as an allegation, not a documented incident.
How the other banks position themselves
In the large-bank segment, the picture is nuanced. Citi CEO Jane Fraser has publicly emphasized governance and control of agent use inside the house: According to Startup Fortune, citing coverage of Fraser's keynote at Sibos 2026 in Miami, the bank has a platform called Arc where no employee can set up an AI agent without routing it through Arc's control and observability layer, so that all agent actions can be traced back to a human owner (Startup Fortune).
At the same time, Crypto Briefing reports, citing the Financial Times, that Citi is also restricting AI agents' access to customer accounts — while Fraser's public statements have centered on strict internal governance rather than an equivalent public blanket ban on external agents (Crypto Briefing). The two sources thus do not give an unambiguous picture of Citi's stance, and the discrepancy between them is unresolved. The safest thing one can say is that both large banks are cautious, but that they communicate it differently.
Interestingly, JPMorgan itself is not holding back on the technology. According to Startup Fortune, the bank, by its own account, has agentic AI running in around 80 internal workflows. The ban therefore does not apply to AI agents as such, but to who is allowed to connect them to customer accounts.
The numbers it all stands against
Dimon's statement came at JPMorgan Chase's third-quarter earnings presentation. According to the company's earnings report filed with the SEC, as reported by Fortune, the bank reported $15.3 billion in net income for the quarter — with $1.6 trillion in average loans and $1.5 trillion in cash and marketable securities — and had total assets of $5.1 trillion as of September 30 (Fortune via Yahoo Finance). In other words: one of the world's largest banks is now explicitly restricting external AI agents in the retail market.
What remains to be answered
Several questions are open. The most important is whether — and when — JPMorgan will allow direct agent interaction; as of now, there is no timeline. Beyond that, it is unclear how liability for agent-initiated payments would be allocated between customer, agent developer and bank if such payments were permitted. Dimon's "liability transfer" concern points to the problem, but no framework resolves it yet.
Ultimately, it also remains to be seen whether other banks follow. With Amazon's block of Muse in September and now JPMorgan's move, much suggests that large platforms and financial institutions will deny external AI agents unrestricted access to their systems — but the picture is still in an early phase, and the Citi question shows that the lines between the players have not yet been fully drawn.
Note: This story rests on secondary reporting of Dimon's remarks at the October 13 earnings presentation and a statement from a Chase spokesperson relayed via CNBC; no bank document is directly available.