This work introduces Axis as a formal legitimacy boundary within the Active Inference framework. While standard Active Inference selects actions by minimizing expected free energy over continuous policy spaces, Axis separates policy optimization from policy admissibility. Policies that violate structural legitimacy constraints are rendered inadmissible, even if they minimize expected free energy. A scalar functional, H3(π), is defined to evaluate structural legitimacy. When no legitimate policy exists, the system defaults to formal abstention rather than forced action. This paper establishes a hard feasibility boundary in policy space, introducing discontinuities that cannot be crossed by gradient-based optimization. Applications include neuroethics, AI safety, and decision-making under existential risk
Darius Hellabad (Sun,) studied this question.