Constitutive AI Unaccountability: A Framework for Systemic Gaps

A recent publication on arXiv (2608.12104) presents the notion of 'constitutive AI unaccountability,' which refers to situations involving various actors, systems, and institutions where achieving AI accountability is fundamentally unfeasible. The authors contend that the current perspective, which views accountability gaps as obstacles to be addressed through standards, transparency, and reforms, falls short. Their research includes a three-phase qualitative study: a literature review focused on concepts, a secondary analysis of 27 interviews with AI experts from technical, legal, and sociotechnical fields, and a case study on the open-source agentic AI system OpenClaw. They delineate nine categories and 20 themes of constitutive AI unaccountability. This framework critiques traditional AI governance methods, positing that some accountability gaps are intrinsic and cannot be rectified merely through better practices. The paper can be accessed at https://arxiv.org/abs/2608.12104.