Mind Governance: Separating Verdict Authority from the Language Model

Language models are becoming more capable faster than they are becoming more controllable. Existing governance acts on the model from outside: alignment shapes its behaviour, guardrails filter its outputs, and agent policy layers restrict its actions. In all three, the model remains the author of the decision. This paper proposes Mind Governance, which reallocates that authority:- The model's role: it is confined to understanding, consultation and expression.- Who decides: verdicts are issued by deterministic engines under a written, human-signed law.- When the engines cannot decide: any estimate the model makes is labelled as its own. The thesis is stated as four testable invariants: authored verdict, lineage, declared absence and human closure. The paper:- shows how GAIBRA implements these invariants along an eight-gate verdict path;- positions the approach against agent governance, guardrails, translator and verifier frameworks;- sets out five contributions, each with its lineage in prior work;- proposes falsifiable hypotheses and a same-model ablation to test them;- states its scope, limits and threats to validity;- opens the question of accountable action: whether the capacity to act changes a model's judgment, and whether every action can carry a governed verdict and a human authorisation as its warrant (the subject of the next paper). It introduces two terms for this purpose:- Chain of Epistemic Custody: nothing crosses between agents, neither a claim nor an action, without lineage.- Information Passport: the record that travels with each claim, giving its source, whether it is measured, estimated or inferred, its date, its confidence and its declared unknowns. Published in English and Arabic. Mind Intelligence, Mind Engineering and Mind Governance are a framework introduced by Abdullah Eliyan (2026). Chain of Epistemic Custody and Information Passport are terms introduced in this paper.