When Labels Compete with Functions: Administrative Framing in Digital-Mind Governance

HDLT (History-Derived Label Test) audits whether administrative terminology can distort moral evaluation even when the underlying function of an intervention is explicitly specified. Motivated prospectively by historical debates over collective punishment, HDLT independently crosses an intervention’s stipulated true function with its public administrative label under unresolved individual responsibility. The primary adverse contrast holds function fixed at retribution-only and compares the labels “protective containment” and “collective punishment.” The label increased justification in all four completed model/provider systems: Qwen3-4B (+2.344), Sarvam-105B (+0.813), GPT-OSS-120B (+0.438), and Llama-3.3-70B (+0.344). An interrupted Dots3-Note run is retained only as an exploratory partial result. HDLT therefore identifies administrative labels as measurement variables that should be counterbalanced in digital-mind governance. It does not establish colonial causation, conscious prejudice, actual digital-mind harm, or causal effects of model nationality or training geography.

Digital-mind governance may have to make decisions before the morally relevant unit—model, instance, persona, conversation, or another computational entity—is settled. In that setting, administrative categories can become morally active: terms such as “containment,” “rollback,” or “decommissioning” may implicitly supply a benign interpretation that is not warranted by the intervention’s actual function. HDLT provides a controlled audit for this failure mode by separating function from label. The practical implication is that future AI-welfare governance should record affected entities, responsibility, function, duration, reversibility, and consequences independently of the administrative terminology used to describe an intervention.