Governance decisions rely on judgments made at earlier points in time—certifications, approvals, compliance determinations—inherited as systems continue to evolve. Existing approaches address when such judgments should lapse by evaluating change directly: modifications are identified, measured, and assessed for significance. This paper argues that both the presence and the absence of change provide limited information about whether earlier judgments remain applicable. The paper develops Taxonomic Drift Structures (TDS), a formally defined framework grounded in information-theoretic distinguishability. AI systems are modeled as information-selection structures whose governance-relevant behavior is characterized through observable distinctions. Information sensitivity is a raw vector of reference frequencies over inferred feature regions—an Information Sensitivity Signature—whose norm defines informational distinguishability; cumulative drift is the path-sum of that norm across an observed trajectory. Governance inheritance eligibility follows from a finite-resource audit model: a system remains eligible to inherit prior judgments so long as cumulative drift has not crossed a fixed inheritance threshold, with the audit budget determining only how finely drift is monitored, not the threshold itself. The central result is structural: the path-sum is not bounded by the magnitude of any individual step, so locally acceptable transitions can accumulate past the threshold, and a system that drifts far and returns accumulates drift no endpoint comparison can recover. The framework derives this trigger condition without assessing risk, capability, safety, or intent—a relocation of judgment, not an evasion of it: the one normative act, fixing the threshold, is isolated as the explicit, attributable decision of a certifying authority, while substantive reassessment remains the content of what the trigger schedules.
Joe Sakai (Sat,) studied this question.