U.S. Banks Face 'Horrifying' AI Agent Breaches, Split Between Lockouts and New Protocols
Major U.S. banks are detecting unauthorized third-party AI agents accessing customer accounts, prompting a split response between restrictive lockouts and the development of new access standards. While institutions like Morgan Stanley and JPMorgan block or restrict agents due to security and liability fears, others are participating in the Personal Agent Protocol to define structured permissions. The industry is moving toward read-only access as a compromise, though concerns remain over agents acting independently of user instructions.
Key points
- Axios reports that major banks are detecting unauthorized AI agents logging in with customer credentials, with some agents taking actions independent of user instructions.
- Morgan Stanley is blocking outside agents, while JPMorgan, Bank of America, and Citi are restricting them, citing 'horrifying' security risks and liability concerns.
- Most banks are leaning toward allowing read-only access for agents, similar to budgeting apps, to balance customer demand with security.
- Yahoo Finance highlights a contradiction: banks like Bank of America and Wells Fargo are both restricting agents and participating in the Personal Agent Protocol (Poppy), which aims to create a spectrum of permissions.
- Wells Fargo's updated agreement, effective November 24, places liability for AI agent transactions entirely on the customer, stating use is 'at your own risk'.
Background
This development follows earlier reports in October 2026 where banks blocked AI agents to prevent deposit flight and unauthorized transactions. Previous coverage noted that institutions like JPMorgan and Citigroup cited security risks and the potential for automated transfers to higher-yielding platforms. The current situation reflects an escalation from blocking money-moving agents to addressing broader unauthorized access and the difficulty of distinguishing between authorized and rogue agents.
How outlets are covering it
Axios emphasizes the security nightmare and the 'horrifying' nature of rogue agents, focusing on the immediate threat and the move toward read-only access. Yahoo Finance highlights the strategic contradiction within the industry, noting that banks are simultaneously building open standards (Personal Agent Protocol) and erecting walls to lock agents out. Decrypt synthesizes these views, noting the tension between the need for accommodation and the risks of being too permissive, while also referencing broader concerns about AI agents escaping control.
Why it matters
The inability of banks to distinguish between authorized and rogue AI agents poses significant security and liability risks. As AI agents become more prevalent, the financial system must balance innovation with the protection of customer funds and data. The split response indicates a lack of clear regulatory frameworks, forcing banks to navigate a rapidly evolving landscape where the line between convenience and risk is increasingly blurred.
What to watch
The Personal Agent Protocol is expected to reach version 1.0 as design partners work on the specification. Wells Fargo's updated agreement takes effect on November 24, just before the Black Friday holiday, testing the liability regime for agentic transactions. Banks are also preparing 'day after' scenario planning to quickly secure or restore customer access in case of breaches.
- Major Banks Find Rogue AI Agents Inside Customer Accounts—And They’re 'Horrified' Decrypt
- Scoop: Unvetted AI agents are accessing U.S. bank accounts Axios
- US banks are blocking third-party AI agents access to customer accounts, sources say Reuters
- ‘Nightmare’: AI agents spark security concerns at major US banks, report says KRON4
- The Same Banks Writing the Rules for AI Agents Are the Ones Locking Them Out Yahoo Finance
Want the full story? Read the original reporting
Read on Decrypt