Abstract
Current debates on AI harm primarily focus on issues directly associated with AI systems, such as algorithmic harm or bias. In this article, I argue that AI harm should be analyzed through a power-aware lens using a systemic and multidimensional approach that accounts for the multiple scales at which harm unfolds—macro, meso, and micro. Reducing AI harm to mere technical failure or a lack of representation in data risks oversimplifying the issue. AI is not just a set of technologies, but a sociotechnical assemblage—a complex interplay of communities, markets, resources, labor, processes, practices, regulations, institutions, and knowledge systems. Its current form not only impacts society at multiple levels but also actively reproduces harm and structural violence at scale, exacerbating power asymmetries both within and across nations. This raises the question of who should be held accountable for harm across multiple scales and what frameworks should be established to address it. A feminist critique of AI frames harm through an ethics of care, emphasizing the need to place human rights at the core of AI governance, ecosystems, and systems.