What Regulators Are Looking For

• Federal banking regulators are questioning banks about autonomous AI agents and how much authority those systems have been given.
• Examiners want to know whether humans remain in control — and whether a bank can immediately shut an AI agent down if something goes wrong.
• The concern goes beyond bad answers from a chatbot. An autonomous system operating inside payments, cybersecurity or risk controls could create consequences before a human recognizes the problem.

Federal regulators are taking a much closer look at something already moving inside the banking system: artificial intelligence that does more than answer questions.

AI agents can act.

The Federal Reserve, Federal Deposit Insurance Corporation and Office of the Comptroller of the Currency are questioning financial institutions about their use of advanced AI, including autonomous agents.

The questions reportedly reach straight into the issue that may matter most — control.

Regulators want to understand what prevents an AI agent from exceeding its authority, whether human beings remain involved in important decisions and whether banks can shut autonomous systems down when necessary. They are also examining cyber risks and the use of foreign-developed AI models.

That distinction matters because an AI agent is fundamentally different from the chatbot many consumers know.

A chatbot may summarize a document. An autonomous agent can potentially receive a goal, decide how to accomplish it, interact with computer systems and perform a sequence of actions with limited human involvement.

Put that inside a bank and the risk changes quickly.

An agent working on cybersecurity could investigate thousands of alerts. One involved in financial-crime compliance could review suspicious transactions. Another might interact with internal systems, customer information or third-party software.

Those capabilities can save enormous amounts of labor. They also create a basic legal and operational question: Who is responsible when the machine acts outside its instructions?

The Federal Reserve has already acknowledged that banks are exploring agentic AI. In March, Fed supervisory official Randall Guynn told Congress that supervisory staff continually monitor banks' use of AI and that institutions were beginning to explore newer generative and agentic systems.

There is also an unusual regulatory gap developing.

Earlier this year, the Fed, OCC and FDIC revised traditional model-risk guidance and specifically said rapidly evolving generative and agentic AI were outside its scope. The OCC nevertheless said the agencies planned further work specifically addressing banks' use of AI, including agentic AI.

So, regulators are now looking closely at technology that existing model guidance deliberately does not fully cover.

That may explain the current fact-finding.

There is no announced federal ban on autonomous AI in banks, and the reported inquiries should not be confused with an enforcement action. Regulators appear to be trying to determine where AI agents are already operating, what permissions they possess, whether their actions can be audited and who can pull the plug.

The stakes are higher than whether an AI system hallucinates a number.

Banks sit at the center of payments, credit, deposits and financial markets. A sufficiently capable autonomous system operating incorrectly — or manipulated through a cyberattack — could execute mistakes at machine speed.

JPMorgan Chase CEO Jamie Dimon raised another side of that problem Tuesday, warning about third-party AI agents obtaining access to customers' bank accounts. JPMorgan does not intend to allow unrestricted access because of the risks such agents could create for both customers and the bank.

The question regulators are confronting is no longer whether banks will use artificial intelligence.

They already do.

The harder question is how much independent authority a machine should ever be allowed to have inside an institution holding other people's money.

About the Author

Samuel López is a reporter and news correspondent for USA Herald with more than two decades of legal-sector experience analyzing litigation, regulatory issues and complex legal matters. His background in law, insurance and compliance informs his reporting on AI governance, institutional accountability and the legal risks created when automated systems begin making consequential decisions.

Keep Following the Stories Behind the Headlines

If you value reporting that goes beyond the press release and examines the legal, financial and regulatory consequences behind major stories, subscribe to the USA Herald newsletter for exclusive reporting and analysis.

👉 Follow USA Herald on X: https://x.com/RealUSAHerald