The Architecture of Authorship: Governing Persistent Human–AI Collaboration

As conversational AI assistants acquire durable cross-session memory and agentic planning capabilities, governing sustained human–AI collaboration raises structural challenges that single-turn benchmarks miss: persistent memory converts sycophancy from a transient conversational annoyance into a state-writing and selective-forgetting hazard (Mao et al., 2026; Pulipaka et al., 2026), while frictionless assistance accumulates cognitive debt that erodes the evaluative substrate required for meaningful human oversight (Gomes Ferreira & Nunes Costa, 2026). This study examines how human authorship, epistemic independence, and final decision authority can be preserved during long-term assistant use through an exploratory longitudinal analytic autoethnographic case study (Anderson, 2006; Kaltenhauser, Stefanidi, & Schöning, 2024) spanning 39 calendar months (approximately 38 elapsed months) across four calendar years (September 2023 – 15 November 2026). Drawing on a partially timestamped archive across technical, creative, commercial, and procedural tasks, the empirical core documents a five-class failure taxonomy—agreement inflation, challenge omission, unwarranted confidence, context/memory failure, and post-rule disclaimer ruptures—alongside a four-axis authorship provenance audit of 12 milestone episodes (GOV-001–GOV-012), the reactive emergence of formal governance rules under live conversational friction, and the qualitative observation of Productivity–Obligation Divergence. Motivated by these failures, the paper formulates a candidate design synthesis (Level E4) comprising a Five-Layer Human Authority Model, a four-stage control sequence (Bound, Filter, Gate, Reset; B → F → G → R), and three prompt-configured interaction registers. Bounded by a single-investigator design (N = 1), pre-October 2025 transcript retention asymmetry, same-day synthesis clustering, and the absence of an unassisted counterfactual, all architectural models, workload equations, and ten candidate design requirements are presented strictly as falsifiable propositions for prospective evaluation rather than validated causal effects.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24323287
Primary Topic
Ethics and Social Impacts of AI
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

The Architecture of Authorship: Governing Persistent Human–AI Collaboration

Brendan T. Henry
Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
preprint

The Architecture of Authorship: Governing Persistent Human–AI Collaboration

Brendan T. Henry
preprint en

Abstract

As conversational AI assistants acquire durable cross-session memory and agentic planning capabilities, governing sustained human–AI collaboration raises structural challenges that single-turn benchmarks miss: persistent memory converts sycophancy from a transient conversational annoyance into a state-writing and selective-forgetting hazard (Mao et al., 2026; Pulipaka et al., 2026), while frictionless assistance accumulates cognitive debt that erodes the evaluative substrate required for meaningful human oversight (Gomes Ferreira & Nunes Costa, 2026). This study examines how human authorship, epistemic independence, and final decision authority can be preserved during long-term assistant use through an exploratory longitudinal analytic autoethnographic case study (Anderson, 2006; Kaltenhauser, Stefanidi, & Schöning, 2024) spanning 39 calendar months (approximately 38 elapsed months) across four calendar years (September 2023 – 15 November 2026). Drawing on a partially timestamped archive across technical, creative, commercial, and procedural tasks, the empirical core documents a five-class failure taxonomy—agreement inflation, challenge omission, unwarranted confidence, context/memory failure, and post-rule disclaimer ruptures—alongside a four-axis authorship provenance audit of 12 milestone episodes (GOV-001–GOV-012), the reactive emergence of formal governance rules under live conversational friction, and the qualitative observation of Productivity–Obligation Divergence. Motivated by these failures, the paper formulates a candidate design synthesis (Level E4) comprising a Five-Layer Human Authority Model, a four-stage control sequence (Bound, Filter, Gate, Reset; B → F → G → R), and three prompt-configured interaction registers. Bounded by a single-investigator design (N = 1), pre-October 2025 transcript retention asymmetry, same-day synthesis clustering, and the absence of an unassisted counterfactual, all architectural models, workload equations, and ten candidate design requirements are presented strictly as falsifiable propositions for prospective evaluation rather than validated causal effects.

Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon