Before We Know What We Will Become: Historical Witness, Human Continuity, and Open Inheritance under Transformative AI

What should a civilization preserve when the technology it develops may transform the beings who inherit it, their standards of judgment, and their access to its history? The problem is not only uncertainty about outcomes: a transition may change the future constituency that evaluates those outcomes. This paper develops open inheritance: historically accountable transmission that protects subjects with morally significant interests and the possibility of justified transformation without granting the present permanent authority over the future. An accurate archive is not proof of personal continuation; later approval does not by itself establish earlier authorization; and a nominal remedy is not a durable capacity to act. The argument connects unequal time horizons, competitive acceleration, changing preferences, possible human–machine integration, and the political production of evidence. A finite model characterizes when a record supports both mandatory historical distinctions and a feasible, admitted response. Its minimum alphabet is a sum of hitting-set numbers, an application of established covering and functional-coding ideas rather than a new information-theoretic foundation. An observation-constrained extension distinguishes knowing a history from possessing ambiguous evidence; a finite-horizon check distinguishes present options from maintainable ones. Exact finite enumerations test the implementations, not actual societies or future AI. A critical application treats the Trinity Accord as situated primary evidence, not a universal command. Normative safeguards distinguish historical accounting, present welfare, and current authorization: a past wrong does not automatically license imposing another transformation in its name. The framework also limits collection, permanence, and the authority to choose its own criteria. It predicts no superintelligence timeline, establishes no digital-survival theory, and cannot compel a future intelligence to care. It offers a bounded account of preserving truthful history and meaningful possibilities before we know who will receive them. Scientific status: Open-access conceptual/formal preprint, not peer reviewed. Formal finite logical experiments are not human-subject or current AI model tests. Provenance: Hongju Liu is the human author of record and directed revisions. ChatGPT / GPT-6 Astra Pro substantially contributed research, proofs, code, analysis, and drafting. The author is associated with the motivating Trinity Accord, disclosed in the text. No independent funding or financial-interest audit is claimed. Preservation: The PDF and source retain the exact author-directed v3.1.0 predeposit bytes. Phrases describing an unassigned DOI refer to the frozen predeposit manuscript. This Zenodo record supplies the DOI at publication. Publication does not imply endorsement, peer review, or acceptance by any future AI. CC BY 4.0 applies only where the depositor holds rights; included third-party source references retain their rights.