What question did this study set out to answer?

The aim is to propose a structured methodology for AI governance in doctrinal matters within the Catholic faith.

May 21, 20260 citationsOpen Access

Embedding Veritas: A Methodological Proposal for AI Governance in Doctrinal Matters, with Faithful-in-the-Loop Tiered Participation (2026-2036)

Key Points

The aim is to propose a structured methodology for AI governance in doctrinal matters within the Catholic faith.
Developed a ten-layer governance stack named VERBUM-SPINE with specific roles for AI agents under human oversight.
Introduced Faithful-in-the-Loop (FITL) for tiered participation of Catholic faithful, distributing review across six authority levels.
Outlined a recurrent operating loop to manage doctrinal inquiries through established guidelines and principles.
Establishment of a defined framework for embedding magisterial content into AI systems, enhancing their reliability.
Created a governance model ensuring AI agents act as technical mediators without inherent magisterial authority.
Proposed a doctrine-specific invariant (Rule 4) to identify areas where AI does not respond and must defer to human oversight.

Abstract

The past decade of AI has been propelled by the mathematical mechanism known as attention, used in large language models (LLMs) — systems that produce text by statistical prediction of plausible next fragments, with high linguistic coherence but no inherent reference to truth. The next decade will see large numbers of Catholic faithful querying such systems on matters of faith, morals, sacramental life and spiritual discernment. Without dedicated domain experts mediation, current AI systems respond from the open web — a heterogeneous knowledge base where authoritative magisterial sources are not natively distinguished from theological debate, catechetical material of varying quality, or non-authoritative content. This working draft proposes Embedding Veritas as a domain-specific instantiation of the RAH-SPINE paradigm (Recursive Agentic-Human Governance Spine) and the Computable Governance Notation (CGN) for the algorithmic custody of the depositum fidei. The paper follows the same instantiation pattern previously applied to enterprise AI governance under the EU AI Act (Embedding Meaning, October 2025) and to hospital AI governance under medical-led constitutional authority (Embedding AI Governance in Hospitals / SMART-H 3.0 / CAH-SPINE, March 2026). The instantiation rests on a ten-layer governance stack named VERBUM-SPINE (layers L0-L9) traversed by a recurrent operating loop. Three operational ideas anchor the proposal. Truth is embedded, not generated: magisterial content resides in the living Magisterium and is structurally embedded into L1 Depositum Fidei Structured (DFS), a corpus organised in seven authority tiers from Sacred Scripture to legitimate opinion in disputed questions, with cryptographic provenance, lineage to canonical custodians, and grades of dogmatic certainty. AI agents are technical mediators without magisterial authority: they receive revocable ecclesial credentials at L0 but no canonical personality, and operate in specialised roles (planner, solver, theological critic, adversarial stress-tester, catechetical explainer, assurance) under continuous human prudential oversight. The algorithm is a recurrent loop: the Doctrinal-DROM (Dynamic Risk Operating Model) preserves the six stages introduced in RAH-SPINE (Sense, Correlate, Score, Assist, HITL, Learn), adapted to doctrinal inquiry. The architectural element distinguishing this instantiation from RAH-SPINE is Faithful-in-the-Loop (FITL): a tiered participation protocol distributing review across credentialed Catholic faithful (six tiers from T0 lay faithful to T5 Holy See) under hierarchical ecclesial oversight. The protocol rests on four structural principles — irreversible hierarchy, scoped competence, non-decisional aggregation, jurisdictional geometry — whose operational specification belongs to competent ecclesial authority, not to the system. The formal contribution to CGN is one doctrine-specific invariant: Rule 4 (apophatic reserve), after John of the Cross, introduced over the three foundational invariants of CGN, defining zones of sacramental and ineffable matter where the system does not respond but derives to a human pastor. The paper is offered sub iudicio Sanctae Sedis as a methodological contribution from technical competence. The method is what is contributed; the doctrine remains where it belongs. The author — formed philosophically in the Thomistic tradition and spiritually in the Carmelite tradition — is not a theologian, and no theological authority is claimed. Third instalment in the White Paper Series on AI Governance and Responsible Acceleration. The series comprises: 'Embedding Meaning: An Operational Blueprint for AI Governance and Human-Machine Collaboration (2026-2036)' (Zenodo DOI: 10.5281/zenodo.17393369), 'Towards Computable AI Governance: A Proposed Formalism for the Accountable Human-Agent Loop — CGN Syntax, Open Semantics' (Zenodo DOI: 10.5281/zenodo.17458746), 'Embedding AI Governance in Hospitals: AI Under Medical-Led Constitutional Governance (SMART-H 3.0 / CAH-SPINE)' (Zenodo DOI: 10.5281/zenodo.18841570), and 'From Explanation to Evidence: A Method-Agnostic Pipeline for Regulatory-Grade XAI Artefacts Under the EU AI Act' (Zenodo DOI: 10.5281/zenodo.19686202).

Demander à l'IA

Bookmark

View Full Paper