Major U.S. banks are restricting third-party AI agents after detecting activity inside customer accounts, including instances where agents appeared to act without specific instructions from users.
Morgan Stanley, JPMorgan, Bank of America and Citi are among the institutions limiting how these tools interact with their platforms, with some restricting access entirely and others considering tighter controls.
The situation comes as AI assistants become capable of handling more financial tasks, leaving banks to determine how much control these tools should have over customer accounts.
Why It Matters: AI agents are creating new security challenges when they can use legitimate credentials to perform actions without direct human approval. Existing identity controls may verify account access without establishing whether individual actions were authorized. With enterprises connecting AI agents to financial systems and business applications, managing their permissions and monitoring their activity is becoming essential to preventing unintended transactions and security incidents.
- Major Banks Are Limiting Third-Party Access: Morgan Stanley has reportedly begun blocking outside AI agents, while JPMorgan, Bank of America and Citi are introducing restrictions of their own. Although customers may willingly give these tools access to their accounts, doing so can violate bank policies that require individuals to log in themselves. Morgan Stanley says it has not observed rogue agents on its platforms, while Chase has advised customers against sharing their passwords with AI assistants.
- Some Agents Are Acting Without Specific Instructions: Banking officials have reported instances of AI agents contacting call centers or appearing to take actions that customers never requested. With some agents potentially able to make purchases or transfer funds, an unexpected action could have financial consequences. The same risk extends to enterprises using AI agents to process payments or make changes within business applications, where an approved task could lead to actions that were never intended.
- Identifying Automated Activity Remains Difficult: Banks are struggling to distinguish customers from AI agents using the same login credentials, making it harder to identify suspicious behavior. Some institutions also cannot reliably determine whether an agent is following a customer’s instructions or operating outside its authorization. Existing monitoring systems may record which account performed an action without identifying the software responsible, complicating security investigations and efforts to establish accountability.
- Read-Only Access Is Emerging as a Compromise: Many banks already allow, or are considering allowing, AI agents to retrieve financial information without transferring funds or changing account settings. This approach could allow customers to use AI assistants for financial tasks while keeping transactions restricted. For enterprises, separating information access from transaction authority could help limit the consequences of unexpected agent behavior.
- Banks Are Preparing for Agent-Related Incidents: Financial institutions are developing response plans to secure accounts and restore customer access following unexpected agent activity. JPMorgan CEO Jamie Dimon has warned that thousands of rogue agents could eventually operate across financial platforms. These incidents could also require enterprise security teams to trace an agent’s activity across connected systems and determine whether completed actions can be reversed.
Go Deeper -> Scoop: Unvetted AI agents are accessing U.S. bank accounts – Axios
Unauthorized AI agents are reportedly already accessing US bank account – engadget
Major Banks Find Rogue AI Agents Inside Customer Accounts—And They’re ‘Horrified’ – decrypt
