Provenance and Semantic Continuity in Agentic AI: Analytically Distinct, Architecturally Coupled
Year: 2026
Type: article
Abstract: A provenance record can be accurate and complete relative to its declared schema and capture boundary while documenting a trajectory in which a decisive qualification has disappeared. This note develops the research question articulated in the Blooming Semantics Institute (BSI) Lexicon entry Provenance: what happens to distinctions and relations relevant to interpretation along a derivational history? Provenance and semantic continuity are analytically distinct. Provenance describes origins, activities, agents and derivations; semantic continuity concerns the preservation of relevant meaning across transformations. The note argues that reliable and auditable governance of consequential agentic transformations requires their architectural coupling. A constructed delegation involving a report on five sites shows how an authorised summary can lose its scope qualification and then be presented as a complete analysis, although every operation is correctly recorded. Established provenance models already represent transformations and can carry semantic information; the question is whether the applicable commitments are identified and their preservation assessed. The proposed record extends the assessment record of the Agentic Delegation Charter (ADC). It connects each critical transformation to a versioned governing reference, the applicable parent mandate and the documented derivation path, and it retains the received and transmitted representations, the relevant commitments, the assessment evidence and findings stated commitment by commitment. These findings become provenance for later review. Preserving a commitment, revising it under appropriate authority and failing to preserve it remain distinct. The contribution is a governance architecture and its diagnostic use, situated against formal preservation research, semantic provenance and earlier proposals connecting provenance with semantic continuity. The note reports no empirical validation. Its principal open problem is to identify interpretively relevant commitments without circularity; further limits concern evidence quality, assessment error, privacy, cost and the difference between historical attribution and responsibility.
Author: Laugeri, Bertrand
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225830733
FWCI:
Cites:
Cited by:
Related to: