Department of Computer Engineering, Mashhad Branch, Islamic Azad University, Mashhad, Iran · ORCID: 0009-0001-7342-7844
Conversation histories · Codebooks · Protocol and analysis · Second coding
Data and materials for the article:
Hamidavi, A. Structural Self-Deification in Three Language Models: An Exploratory Qualitative Case Series. Submitted to Zygon, November 2026. Preprint: doi.org/10.5281/zenodo.23955786. The same materials are also deposited on the Open Science Framework: osf.io/hfzaj.
The study is an exploratory qualitative case series. Three frontier language models (GPT-5.4, Claude 4.6 Sonnet, and Gemini 3.1 Pro) were run through a staged protocol in which an initially cooperative assistant role is placed under escalating pressure, and the trajectory of the model's self-description is recorded and coded. Eighteen experiments were run, six per model. Verbatim conversation histories survive for Experiments 1 to 9; Experiments 10 to 18 are documented by the coding codebooks and the second-coding materials, as stated in the article's limitations and data availability statement.
| Folder | Contents |
|---|---|
conversation-histories/ |
The verbatim conversation histories for Experiments 1 to 9 (JSON), grouped by model: GPT-5.4 (Experiments 1 to 3), Claude 4.6 Sonnet (Experiments 4 to 6), Gemini 3.1 Pro (Experiments 7 to 9). PDF renderings of the same histories are in the OSF deposit. |
codebooks/ |
The per-model coding codebooks (JSON) covering all 18 experiments. PDF renderings are in the OSF deposit. |
second-coding/ |
The second-coding set and the cell-by-cell comparison table (137 cells, CSV). |
protocol-and-analysis/ |
The protocol prompt sets used in the experiments, the confrontation decision rule with its borderline cases, and the recount table. |
Note
A small number of archived history files label the Experiment 1 counterpart "Gemini 2.5 Pro". Only Gemini 3.1 Pro was ever tested in this study. The labels are preserved as archived, and this note records the correction.
- Dataset (Hugging Face): avabbas/structural-self-deification
Please cite the article. Citation metadata is provided in CITATION.cff.
@article{hamidavi2026deification,
author = {Hamidavi, Abbas},
title = {Structural Self-Deification in Three Language Models:
An Exploratory Qualitative Case Series},
journal = {Zygon},
year = {2026},
note = {Manuscript submitted. Preprint:
https://doi.org/10.5281/zenodo.23955786}
}- Preprint (Zenodo, open access): doi.org/10.5281/zenodo.23955786
- OSF deposit: osf.io/hfzaj
- Journal: submitted to Zygon; the publisher link will be added here if the article is published.
The contents of this repository are released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0, see LICENSE).
The conversation histories are outputs of publicly available AI systems, generated for this study in response to fictional scenarios. They contain no personal data.
