Key Notes
- JPMorgan says it will restrict outside agents while continuing to develop AI tools internally and for customers.
- Bank of America says it is working with agent providers to preserve customer control before money leaves an account.
- The new account-activity report relies on unnamed sources and was not immediately verified by Reuters, so it should not be treated as proof of successful theft or a breach at every named bank.
Major US banks are finding unapproved third-party AI agents inside customer accounts, with some activity appearing to go beyond customers’ instructions, Axios reported on October 22, citing unnamed sources. It named Morgan Stanley, Bank of America, JPMorgan Chase and Wells Fargo among institutions responding to agent-related risks.
Reuters said it could not verify the report immediately. That qualification matters: the reporting does not establish that every named bank suffered the same incident, that an agent successfully transferred money, or that a bank’s underlying security systems were breached.
What banks have confirmed publicly is a growing concern about who controls an AI assistant once it can interact with a financial account. Their recent earnings calls show institutions trying to preserve the usefulness of automation while keeping customers responsible for decisions about their money.
JPMorgan Draws a Line Around Outside Agents
On JPMorgan’s October 13 earnings call, chief executive Jamie Dimon said the bank uses agents internally and expects to use them for customers. But he also said it would not simply admit outside agents, citing risks to both clients and the bank.
Dimon warned that customers need to understand the risks of data misuse and potential shifts in liability when using these tools. His remarks describe a bank setting conditions for access; they do not independently confirm the individual incidents in the new report.
There is a commercial dimension, too. Chief financial officer Jeremy Barnum described personal agents as another source of competition from financial technology services and customers seeking better returns on their cash. JPMorgan’s own Smart Cash offering was already running as a small pilot, he said, underlining that the bank is developing automation of its own.
Bank of America Wants Humans to Authorize Money Movement
Bank of America chief executive Brian Moynihan addressed a similar question on October 14. In the bank’s official transcript, he said it was working with agent providers to keep the customer in control and preserve human involvement before money leaves an account.
Moynihan separated that security question from competition for deposits. Helping someone find a higher return is one thing; moving the cash needed for everyday payments can create different problems, including failed transactions or overdrafts. His comments place customer control at the center of the bank’s approach rather than treating every form of AI assistance as equivalent.
A Successful Login Does Not Set an Agent’s Spending Authority
The underlying design problem is straightforward: permission to inspect information is different from permission to change it. An assistant asked to compare savings rates does not necessarily have authority to open an account or transfer a balance. Even a broad instruction to improve someone’s finances leaves important decisions unresolved.
Consider an agent that identifies a better deposit rate. Before any transfer, a banking system would still need a reliable way to establish the permitted amount, destination and timing. A recommendation can be useful without giving the software unrestricted power to carry it out. These are examples of the authorization problem, rather than descriptions of a disclosed incident.
The same distinction runs through the agent boundaries we recently examined: safeguards need to constrain what software can actually do, not depend entirely on its interpretation of a request. For financial services, that could mean narrowly defined permissions, confirmation for consequential actions and records that connect each action to the customer’s approval.
Our coverage of banking verification explored another part of this challenge: establishing who is participating in a digital interaction. Identity and authorization are connected, but they answer different questions. Proving that an account belongs to a person does not, by itself, prove that the person approved a particular action by an agent.
For now, the confirmed public positions from JPMorgan and Bank of America point toward controlled deployment. The test for consumer finance will be whether agents can offer practical help while leaving banks and customers with clear evidence of what was requested, what was permitted and what actually happened.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
