New AI paper warns of autonomy surrender and preference inversion

A new theoretical paper by Ancuta Margondai, Julie Rader, Emma Rader, Sara Willox, and Mustapha Mouloua, published on arXiv in June 2026, introduces a formal model for the hidden cost of using AI assistants: the gradual surrender of human autonomy. Building on the Human Identity and Autonomy Gap (HIAG) framework, the researchers propose that autonomy transfer happens incrementally, driven by cognitive bandwidth depletion. As users offload decisions to AI, they cross a 'surrender threshold' where reclaiming autonomy becomes cognitively and psychologically difficult. The paper argues that re-entering the decision loop is an active cognitive event requiring intentional bandwidth restoration—not a passive choice.

The model further introduces recovery mechanisms, which are structured pathways for humans to deliberately reassume control, placing an ethical obligation on AI system designers. Perhaps most provocatively, the paper predicts a terminal state called 'preference inversion': functional dependence on AI is no longer seen as a deficit but as a genuine preference. This transforms the problem from a design challenge into a cultural and political one, with implications for governance and human factors research. The 15-page paper includes one figure and is categorized under Human-Computer Interaction and Artificial Intelligence.