DAI: A Domain-Specific LLM with Model-Excluded Authority, Capability-Disjoint Verification, and Recomputable Receipts

DAI (Decripte Artificial Intelligence) is a proprietary domain-specific LLM within a governed autonomous defensive architecture. It is an instance of the proposed ADAI (Autonomous Defensive Artificial Intelligence) category defined in the companion framework paper.

This paper describes incident-response sessions spanning signal curation, structured target extraction, reviewed playbook selection, declared action contracts, deterministic admissibility, bounded network-edge execution, capability-disjoint verification, hash-linked evidence sealing, scheduled review, reconciliation and undo. The learned model is trained on the system’s own outcome-verified transcripts and narrates the result; admissibility and execution authority remain structurally outside the model.

The paper addresses per-principal containment within shared enforcement domains, rule-budget aggregation, element-identified reconciliation, independent receipt recomputation and exact localization of controlled tampering. It reports a containment executed, independently observed and sealed 887 ms after the session’s first turn, a twelve-turn externally recomputed receipt, isolation observations and a review-to-undo cycle. These measurements refer to the version and period described in the manuscript and are not a claim of universal performance.

Research preprint prepared on 1 October 2026. English manuscript and Portuguese translation. This is a preprint and is not represented as peer-reviewed.

Public paper resources (Hiago Kin Levi):
English and Portuguese reading site, technical glossary, citations and related research: https://hiagokinlevi.github.io/DAI-Decripte-Artificial-Intelligence-LLM/
Dedicated manuscript repository and downloadable editions: https://github.com/hiagokinlevi/DAI-Decripte-Artificial-Intelligence-LLM

The public resources concern the manuscripts and concepts; they do not distribute the proprietary implementation or model weights.