Dimensional characterization and Pathway Modeling for Catastrophic AI Risks

This paper examines catastrophic AI risks using a framework that combines seven risk dimensions: intent, competency, entity, polarity, linearity, reach, and order, along with risk pathway modelling to link hazards to harms. It analyzes six major risk types: Chemical, Biological, Radiological, and Nuclear (CBRN); cyber offence; sudden and gradual loss of control; environmental; and geopolitical. The paper outlines the risks dimensions, causal pathways, and mitigation measures, providing a structured foundation for managing catastrophic AI risks across the value chain