The COST of AI: A Multi-Level Framework of AI Adoption for Malevolent Innovation

Abstract Drawing on the malevolent creativity and innovation literature, the creative problem-solving process, and diffusion of innovation (DOI) theory, the present effort examines the conditions under which malevolent actors may adopt artificial intelligence (AI) in pursuit of harmful objectives. We present the Cognitive-Operational-Societal- Technology (COST) model, a theoretical framework that integrates and organizes AI-enabled malevolent innovation across three distinct levels. At the cognitive level, we postulate that AI serves to reduce expertise constraints and expand the population of actors capable of harmful ideation. At the operational level, we propose that AI narrows the distance between intent and execution by reducing the psychological and technical barriers to implementation. At the societal level, AI’s persuasion characteristics may accelerate the diffusion of harmful capability. Following these descriptions, several cases are used to illustrate differing applications of AI for malevolent use. This effort pursues three core aims: examining how AI differs from prior technologies in malevolent use, applying DOI theory to explain its perceived advantage for malevolent actors, and introducing the COST model to organize AI-enabled harms across multiple levels.