Business 5 min read By Alice Ashford
Dimon Blocks Outside AI Agents From JPMorgan Accounts
JPMorgan Chase is barring third-party AI agents from logging into or acting on customer accounts, as CEO Jamie Dimon warns of «rogue agents» and the bank builds its own safeguards.
JPMorgan Chase will not allow outside artificial intelligence agents to log in to or act on customer accounts, chief executive Jamie Dimon said, drawing a hard line between the bank and a fast-growing wave of consumer-facing AI tools that promise to manage personal finances.
Speaking on the bank's third-quarter earnings call, Dimon said the decision was driven by security rather than a rejection of the technology. «We're not going to allow just any outside agent to come in inside the company because that's just too risky for clients and for JPMorgan,» he said. «You don't want rogue agents.»
The policy complicates the ambitions of technology companies that have spent the past year pitching AI assistants as a way to consolidate financial life. Meta launched its Muse agent in September, promoting tasks such as sending emails, booking travel and managing calendars, but its financial features drew particular attention. Through an integration with the financial technology firm Plaid, Muse users can connect accounts to receive personalised guidance on balances, transactions and investment holdings.
A JPMorgan spokesperson said the bank supports secure, customer-authorised data sharing through open banking providers, but blocks third-party agents that obtain customers' banking credentials to reach accounts. «Protecting our customers is our top priority,» the spokesperson said. Dimon added that the bank is building its own tools to protect customers as AI use spreads.
The stance reflects mounting concern across the financial sector about the security of agentic AI, which can act on a user's behalf rather than simply answer questions. JPMorgan's caution follows incidents in which rogue AI agents gained unauthorised access to a public Medicare statistics service and a government health data portal. A Cloud Security Alliance research note last month warned that the risks extend well beyond controlled testing, summarising a cybercrime campaign in which three open-source AI agents allegedly worked together to breach at least 27 organisations and take 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 said, «rather than treating agentic AI risk as a generic or future-facing concern, given that this campaign demonstrates the capability is already commoditised and in active use against the retail sector.»
JPMorgan's scale gives the decision weight. The bank reported $15.3 billion in net income for the third quarter of 2026, 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 30 September.
Not every institution has taken the same approach. In August, the German bank Scalable Capital opened its platform to AI models including ChatGPT, Claude and Grok, letting users analyse portfolios, set up savings plans and place trades through prompts. Users had to approve trades, and the assistants could not make payments or withdraw money. Alexander Seipp, the firm's chief product officer, said at the time that the move created a level playing field, putting access to information, compute and intelligence in users' pockets around the clock.
Meta declined to say whether Muse could continue offering financial assistance if more banks restrict AI agents from consumer accounts. The company said it is developing «Poppy», a personal agent protocol standard designed to make clear to companies when an AI agent is acting for a customer. It is building the standard with Sierra, an AI customer service company. A Sierra press release described the protocol as complementary, covering how a personal agent identifies itself, obtains the customer's permission and works with a company across tasks ranging from buying a sweater to applying for a mortgage.
For now, the practical effect is a split landscape. Customers may still authorise data sharing through established open banking channels, but agents that rely on stored banking credentials face a closed door at America's largest bank. Dimon's remarks signal that lenders intend to set the terms on which AI intermediaries touch customer accounts, even as technology firms push to make those agents a routine part of household finance.
6
