"I Am the Gatekeeper": Clinician Visions of AI Collaboration in Exposure Therapy

Abstract

Exposure therapy is one of the most effective treatments for anxiety and trauma-related disorders, yet many trained clinicians avoid it in practice. Its delivery is cognitively, relationally, and creatively demanding in ways that existing AI tools have not addressed. Research on AI in mental healthcare has focused on evaluative systems that score, flag, and recommend, leaving clinicians as consumers of AI outputs rather than partners in clinical reasoning. To understand how clinicians themselves envision AI collaboration, we conducted semi-structured interviews with twelve licensed exposure therapy specialists and analyzed the data through reflexive thematic analysis. We identified four attributes of the ideal AI teammate: a silent in-session observer, a creative partner in hierarchy design, a transparent reasoner, and a collaborator whose autonomy expands incrementally under clinician gatekeeping. We discuss how these attributes form a coherent relational model of human-AI collaboration, which we term relational AI architecture, and conclude with design and research implications for clinical AI.