This preprint introduces a minimal and testable condition for identifying the transition from response to commitment in AI-mediated decision processes. Rather than locating this boundary at execution or irreversibility, the paper proposes the loss of neutral return as the decisive threshold: the point at which a system can no longer proceed as if a given output had not occurred. The condition is developed through a controlled boundary case in which the underlying facts remain constant while a change in framing alters the procedural status of the model’s output. This allows the transition to be observed independently of new information or external intervention. The resulting framework distinguishes between formal reversibility and neutral return, showing how systems may remain technically reversible while already operating under altered decision conditions. The proposed condition is invariant to execution, visibility, and system scale, providing a minimal surface for analyzing admissibility, commitment, and irreversibility across AI-mediated and agentic systems.
Skulski et al. (Thu,) studied this question.