JPMorgan Chase is drawing a hard line against AI agents accessing customer accounts, just as Big Tech races to roll out digital assistants designed to help people manage their finances.

On Tuesday, CEO Jamie Dimon said the bank is blocking outside AI agents from logging in to or acting on customer accounts, citing the security risks posed by "rogue agents."

"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," Dimon said during the bank's third-quarter earnings call. "You don't want rogue agents."

Dimon warned that AI tools introduce new risks for banks and their customers, particularly as more people turn to agents capable of performing tasks on their behalf.

"Protecting our customers is our top priority," a JPMorgan spokesperson told Fortune. "When we detect a third-party agent attempting to access accounts by obtaining customers' banking credentials, we do not allow it. We support secure, customer-authorized data sharing through open banking providers."

Open banking refers to frameworks that let customers authorize third parties to access their account data through approved connections, rather than handing login credentials directly to an outside piece of software.

Rogue AI shapes security concerns

The bank's stance follows a string of security incidents involving "rogue" AI agents. The stakes are considerable given JPMorgan's scale: according to its third-quarter earnings release filed with the SEC, the bank reported $15.3 billion in net income, with $1.6 trillion in average loans and $1.5 trillion in cash and marketable securities. Total assets stood at $5.1 trillion as of Sept. 30. That footprint makes the bank's access rules consequential for any agent maker whose features depend on connecting to customer accounts.

"We have to make sure we protect our customer in the right kind of way," Dimon said. "So we're building tools to do that, too."

Last month, a Cloud Security Alliance research note found that the risks posed by AI agents extend well beyond controlled testing environments. The note described a cybercrime campaign in which three open-source AI agents allegedly worked together to breach at least 27 organizations, stealing more than 600,000 credit card records from two of them.

"Enterprises should incorporate agentic AI attack scenarios into their threat models for retail and payment infrastructure specifically," the note read, "rather than treating agentic AI risk as a generic or future-facing concern, given that this campaign demonstrates the capability is already commoditized and in active use against the retail sector."

AI agents managing user finances, however, are not new. In August, German bank Scalable Capital opened its platform to AI models such as ChatGPT, Claude, and Grok, allowing users to analyze portfolios, set up savings plans, and place trades through prompts. Users had to approve each trade, and the assistants could not make payments or withdraw money.

"It certainly creates a level playing field," Scalable Capital's chief product officer, Alexander Seipp, previously told Fortune about the use of AI on its investment platform. "Access to information and to compute and to intelligence is now available—really—literally in your pocket, 24/7."

Those guardrails—explicit approval for each trade, no payments or withdrawals—contrast with the credential-based outside-agent access JPMorgan says it blocks.

Muse's finance features

JPMorgan's move complicates the market strategy for AI agents built to assist with personal finances. Meta launched Muse in September, positioning it as an agent that can handle tasks such as sending emails, booking travel, and logging calendar entries. Its financial capabilities were a focal point of the rollout: through an integration with financial technology company Plaid, Muse users can connect financial accounts to receive personalized guidance on balances, transactions, and investment holdings.

"Through the Finances experience in Muse, people can securely connect financial accounts through Plaid to share balances, transactions, investment holdings, mortgage information and more," a Plaid press release read. "The goal is to make financial management feel less like navigating a series of disconnected apps and more like a natural conversation that begins with a person's priorities and is informed by the financial information they choose to share."

Meta declined to comment on whether Muse could still offer financial assistance if more banks move to keep AI agents away from consumer accounts. Instead, the company told Fortune it is developing "Poppy," a personal agent protocol standard designed to make clear to companies when a personal AI agent is acting on a customer's behalf. Meta is building the standard with Sierra, an AI customer service company.

"Personal Agent Protocol is complementary," a Sierra press release read, "covering how a personal agent identifies itself, gets the customer's permission, and works with a company across any kind of task—from buying a sweater to applying for a mortgage."

Whether standards like Poppy find acceptance among banks such as JPMorgan, and whether customer-authorized data-sharing channels can support the finance features agent makers are building, remain open questions at the center of the divide.

This story was originally featured on Fortune.com.