Contemporary AI governance frameworks predominantly classify risk at the level of model outputs, system capabilities, or post-hoc harms. Across prior analyses of professional liability, laboratory evaluation limits, assurance gaps, and epistemic boundary drift (fragmentation in how responsibility and oversight are conceptualised across domains), a consistent structural misclassification has emerged: AI-mediated language interaction participates in human judgment formation prior to decision execution, yet remains formally external to recognised governance surfaces. This paper consolidates those analyses by identifying AI-mediated judgment interaction as a distinct governance surface that is currently unclassified. The argument is diagnostic rather than prescriptive. It does not propose regulatory reform, standards modification, or technical intervention. It establishes a classification clarification necessary for institutional alignment across governance, assurance, laboratory evaluation, and academic domains.
Victoria Gavaza (Tue,) studied this question.