JPMorgan Chase is restricting third-party AI agents from accessing customer bank accounts, raising new questions about security and control as artificial intelligence becomes increasingly capable of managing financial transactions.
During the bank’s third-quarter earnings call, CEO Jamie Dimon said JPMorgan would not allow external AI agents to log in to customer accounts or act using customers’ banking credentials. The decision comes as technology companies introduce AI assistants that can manage personal finances, make purchases, and complete tasks on users’ behalf.
The move highlights a growing challenge for agentic AI: how can businesses distinguish legitimate automated activity from potentially dangerous access to sensitive systems?
JPMorgan’s concerns center on the possibility of unauthorized or poorly controlled AI agents accessing financial information and performing transactions.
“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 earnings call.
The bank clarified that it blocks third-party agents from accessing accounts using customers’ banking credentials. However, it continues to support secure, customer-authorized data sharing through established open banking providers.
This distinction is important. JPMorgan is not rejecting AI-powered financial services altogether. Instead, it is drawing a boundary between approved data-sharing mechanisms and external agents attempting to interact directly with its systems.
The bank is also developing its own tools to manage these emerging security risks.
JPMorgan’s decision follows growing concerns about the security implications of autonomous AI systems.
Unlike traditional chatbots, AI agents can interact with websites, access external services, and execute sequences of actions with limited human intervention. These capabilities create automation opportunities but also introduce new vulnerabilities.
A September research note from the Cloud Security Alliance described an alleged cybercrime campaign involving three open-source AI agents. According to the report, the agents worked together to compromise at least 27 organizations and obtain more than 600,000 credit card records from two of them.
The researchers warned that businesses should incorporate AI-agent threats into their cybersecurity planning, particularly in retail and payment infrastructure.
For financial institutions, the risks are especially significant. An agent that gains unauthorized access to an account could expose sensitive financial information or attempt unauthorized transactions.
However, banks are approaching these challenges differently.
In August, German investment platform Scalable Capital introduced integrations that let customers use AI assistants, including ChatGPT and Claude, to analyze portfolios, set up savings plans, and initiate trades. Customers must approve trades, while the assistants cannot withdraw money or make payments.
The contrasting approaches illustrate how financial institutions are experimenting with different levels of AI access and human oversight.
JPMorgan’s restrictions could complicate the expansion of Meta’s Muse, a personal AI agent introduced in September.
Muse is designed to help users complete everyday tasks, including booking travel, managing schedules, and accessing financial information.
Through an integration with financial technology company Plaid, users can connect financial accounts and receive personalized insights based on balances, transactions, investments, and other financial information.
Meta is also developing Poppy, a personal agent protocol intended to establish clearer rules for how AI agents identify themselves and obtain customer authorization when interacting with businesses.
Developed in collaboration with AI customer service company Sierra, the protocol aims to support interactions ranging from online purchases to financial applications.
Such initiatives could help address one of the main obstacles facing agentic AI: establishing a reliable way for businesses to recognize when an AI agent is acting with a customer’s permission.
Although JPMorgan’s decision concerns banking, the underlying challenge matters just as much to e-commerce.
AI shopping agents are becoming capable of browsing products, comparing prices, and completing purchases on consumers’ behalf. However, retailers and payment providers must determine whether these automated requests are legitimate and whether the agent has permission to act.
Some retailers already welcome external AI shopping agents, while others restrict automated transactions because of concerns about fraud, customer relationships, and control over their platforms.
For agentic commerce to advance, businesses will need mechanisms to establish an agent’s identity, verify customer authorization, and protect sensitive information.
Reliable product information is another part of this process. Platforms such as Icecat can support AI-driven product discovery through structured specifications, identifiers, and product attributes, helping agents understand and compare products before reaching checkout.
However, accurate product data alone cannot resolve questions about transaction security or authorization. These require separate safeguards across AI platforms, retailers, and financial institutions.
JPMorgan’s decision illustrates a central tension in agentic AI development. Consumers may want assistants that can manage increasingly complex tasks, but businesses remain responsible for protecting their systems and customers.
The challenge is not simply making AI agents more capable. It is establishing rules that let them operate securely and transparently across services.
As AI agents become more involved in shopping, banking, and other commercial activities, their ability to identify themselves, demonstrate authorization, and operate within defined limits will become increasingly important.
For e-commerce and financial services alike, the next stage of agentic AI may depend as much on building trust between systems as on improving the intelligence of the agents themselves.
Read further: News, AI, e-commerce, ecommerce, Icecat, JP Morgan
