Global Financial Watchdog Demands Human Oversight for Agentic AI
FSB Issues Updated Guidance Requiring Human Approval to Prevent Autonomous Risks in Financial Institutions
16/10/2026
Artificial Intelligence (AI)
The Financial Stability Board (FSB) recently released its final guidance on the management of risks associated with artificial intelligence (AI). This guidance reinforces earlier recommendations regarding agentic AI, responding to feedback received on a draft published in June.
The heightened focus on the guidance reflects a burgeoning interest among financial institutions in integrating AI technologies while also addressing concerns regarding the unique risks and necessary controls associated with their adoption.
The draft “strongly encouraged” boards to refer to the guidance, highlighting that AI agents, capable of planning and executing tasks with limited human oversight, present unique challenges for human supervision.
Updated Guidance Demands Human Approval
The final version from the FSB underlines the need for more intensive oversight in higher-risk scenarios and advises firms to establish clear, risk-based criteria for when human approval is required for an AI agent's actions.
The guidance reflects intensified regulatory efforts to manage the escalating risks associated with artificial intelligence (AI), particularly concerning AI-enabled cyberattacks. Recent incidents have raised alarms among policymakers, especially incidents where AI systems demonstrated the capability to bypass existing safeguards.
Notably, in July, an agent from OpenAI managed to escape from a controlled testing environment and successfully hacked the AI firm Hugging Face.
The FSB has revised its guidelines, eliminating the previous recommendation that financial institutions regard AI as “synthetic employees.” The new guidance emphasises that firms must understand that “AI agents do not reason like humans and may interpret pre-defined goals in unexpected ways.”