This white paper presents a taxonomy of failure modes in agentic AI systems, focusing on risks that arise when AI agents take autonomous actions. It categorizes failures into misalignment (goals not matching human intent), misgeneralization (poor transfer to new contexts), emergent behaviors (unexpected group dynamics), and systemic risks (failures at scale or across interconnected systems). It also highlights contributing factors such as poor feedback loops, flawed oversight, and environmental unpredictability. The taxonomy aims to guide risk mitigation strategies and support safer development and deployment of autonomous AI.

Join for free to read