Governance in Agentic Workflows: Leveraging LLMs as Oversight Agents
Keywords: Agentic, Governance, Judge, Fairness, Transparency
Abstract: Agentic workflows, composed of interconnected AI agents performing complex tasks, have become extremely popular both in the academic and industrial communities. While such workflows possess significant potential to help processes become more efficient and targeted, they also present risks. As an increasing number of agentic workflows are being deployed in real-world environments, a holistic governance approach is necessary to ensure accountability, safety, and traceability. This paper proposes a Governance Judge which can be incorporated to modern day workflows. By addressing practical issues such as performance monitoring, failure detection, and compliance auditing, this approach helps ensure robust oversight for agentic workflows in dynamic operational contexts.
Submission Number: 6
Loading