Agentic AI has a Human Oversight Problem

Synopsis

Position paper arguing effective human oversight of agentic AI is currently infeasible because of a goal-plan-execution gap, observation difficulty, and the contextual auditing nature of agent decisions.

Keywords

·agentic AI ·human oversight ·sociotechnical ·position paper ·governance

Related entries

• AGENTSAFE: A Unified Framework for Ethical Assurance and Governance in Agentic AI
  December 2, 2025 · arXiv
• The AI Agent Index
  February 3, 2025 · arXiv
• Position: Towards Bidirectional Human-AI Alignment
  June 13, 2024 · ICML 2024
• On the Quest for Effectiveness in Human Oversight: Interdisciplinary Perspectives
  April 5, 2024 · FAccT 2024
• Practices for Governing Agentic AI Systems
  December 14, 2023 · OpenAI white paper 2023
• Agentic AI in the Software Development Lifecycle
  April 29, 2026 · arXiv