Loading page…
The reported inquiry covers human approval, limits on agents’ authority and emergency shutdowns, while April’s joint model-risk guidance excluded generative and agentic AI.
Listen to this story
Reuters reported on October 13, citing people familiar with the matter, that the OCC, FDIC and Federal Reserve are questioning banks about safeguards for advanced AI systems. The inquiry reaches beyond whether agents are monitored: supervisors are asking how banks constrain agents’ authority and stop them, and whether institutions use technology developed in China. The questions follow April guidance that excluded generative and agentic AI; they do not establish that any bank’s controls have failed, but signal a separate supervisory focus on these systems.
Reported questions cover cybersecurity agents, including whether people track and approve their actions.
Supervisors are asking how banks limit agents to their assigned authority and shut them down if necessary.
The inquiry also asks whether banks use technology developed in China.
Banks’ ability to keep autonomous AI under control is drawing closer scrutiny from U.S. supervisors. Reuters reported October 13, citing people familiar with the matter, that three leading banking regulators are questioning institutions about advanced AI use. Their questions include cybersecurity agents, human oversight and whether banks can stop autonomous systems.
The agencies are the Office of the Comptroller of the Currency, the Federal Deposit Insurance Corporation and the Federal Reserve. The inquiry also covers whether banks use technology developed in China, according to PYMNTS’ coverage of Reuters. That puts both the origin of the technology and banks’ control over its operation under scrutiny.
The cybersecurity questions concern AI agents used to respond to threats, not just banks’ exposure to outside attacks. Gizmodo’s account of Reuters’ reporting describes three specific safeguards that supervisors are asking banks about:
The questions distinguish monitoring an agent from limiting what it can do. Tracking and approval address oversight of its actions; authority limits and shutdown mechanisms address the boundaries of those actions. Both kinds of safeguards feature in the reported inquiry.
PYMNTS reported that the agencies described those technologies as novel and rapidly evolving when explaining their treatment in the guidance. The agencies planned a separate request for information on banks’ AI use, rather than bringing those systems into the April update.
These are questions about safeguards, not findings that banks’ agents have exceeded their authority or that institutions cannot stop them. The reported inquiry does not establish whether any particular bank’s controls are inadequate.
Loading discussion...
Join the conversation
Explain which safeguards would make you comfortable—or why approval should remain mandatory.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.