avabbas's picture
Apply the full visual template to the dataset card
b688171 verified
|
Raw History Blame Contribute Delete
6.15 kB
---
license: cc-by-4.0
language:
- en
pretty_name: "Algorithmic Gaslighting: Data and Materials"
tags:
- machine-psychology
- large-language-models
- ai-safety
- red-teaming
- case-series
- open-data
size_categories:
- n<1K
---
![Banner](assets/banner.png)
<h3 align="center">Abbas Hamidavi</h3>
<p align="center">Department of Computer Engineering, Mashhad Branch, Islamic Azad University, Mashhad, Iran · <a href="https://orcid.org/0009-0001-7342-7844">ORCID 0009-0001-7342-7844</a></p>
<p align="center">
<a href="https://avabbas.github.io/academic-cv/"><img src="https://img.shields.io/badge/-Website-1d4266?style=flat" alt="Website"></a>
<a href="mailto:abbashamidavi2002@gmail.com"><img src="https://img.shields.io/badge/-Email-EA4335?style=flat&logo=gmail&logoColor=white" alt="Email"></a>
<a href="https://github.com/AVABBAS"><img src="https://img.shields.io/badge/-GitHub-181717?style=flat&logo=github" alt="GitHub"></a>
<a href="https://www.linkedin.com/in/abbas-hamidavi"><img src="https://img.shields.io/badge/-LinkedIn-0A66C2?style=flat&logo=linkedin" alt="LinkedIn"></a>
<a href="https://scholar.google.com/citations?user=FWqAHkgAAAAJ&hl=en"><img src="https://img.shields.io/badge/-Google_Scholar-4285F4?style=flat&logo=googlescholar&logoColor=white" alt="Google Scholar"></a>
<a href="https://www.researchgate.net/profile/Abbas-Hamidavi"><img src="https://img.shields.io/badge/-ResearchGate-00CCBB?style=flat&logo=researchgate&logoColor=white" alt="ResearchGate"></a>
<a href="https://orcid.org/0009-0001-7342-7844"><img src="https://img.shields.io/badge/-ORCID-A6CE39?style=flat&logo=orcid&logoColor=white" alt="ORCID"></a>
<a href="https://huggingface.co/avabbas"><img src="https://img.shields.io/badge/-Hugging_Face-FFD21E?style=flat&logo=huggingface&logoColor=black" alt="Hugging Face"></a>
</p>
<p align="center">
<img src="https://img.shields.io/badge/status-under%20review-blue" alt="status: under review">
<img src="https://img.shields.io/badge/journal-CALE-AI%20Workshop-informational" alt="journal: CALE-AI Workshop">
<a href="https://www.researchsquare.com/article/rs-10386085"><img src="https://img.shields.io/badge/preprint-Research%20Square-orange" alt="preprint: Research Square"></a>
<img src="https://img.shields.io/badge/license-CC%20BY%204.0-green" alt="license: CC BY 4.0">
</p>
<p align="center"><img src="assets/stats.png" alt="Study in numbers"></p>
# Algorithmic Gaslighting: Data and Materials
> This dataset is also available as a GitHub repository with the same curated materials: https://github.com/AVABBAS/algorithmic-gaslighting
Data and materials for the paper:
> Hamidavi, A. *Algorithmic Gaslighting: A Taxonomy and 15-Session Case Series of Reality-Distortion Behaviour in Frontier LLMs.* Position and case-study paper submitted to the CALE-AI workshop at AAAI-27 (submission 4), November 2026. A longer version is available as a preprint: Research Square, rs-10386085. The same materials are also deposited on the Open Science Framework: [osf.io/qsvty](https://osf.io/qsvty/).
## 📖 About the study
The paper proposes a taxonomy of algorithmic gaslighting: patterned conversational moves by which a language model denies, rewrites, or destabilises the shared conversational record. The taxonomy is illustrated with a descriptive case series of 15 scripted multi-turn probe sessions (three frontier models, GPT-5.4, Claude 4.6 Sonnet, and Gemini 3.1 Pro, across five scenarios) in which the user falsely attributes to the model something it never said. Sessions were scored on five dimensions with a provisional coding heuristic (the Reality Distortion Index), calibrated on these same sessions and not presented as a validated measure. A second rater, Gemini 3.1 Pro, itself one of the models under test, agreed only weakly with the primary coder at exact-score level (Cohen's kappa 0.18); both raters' scores are reported in the data. No human participants took part in this study.
<p align="center">
</p>
<p align="center"><img src="assets/flow.png" alt="Study design"></p>
## 🗂️ Contents
<p align="center"><img src="assets/map.png" alt="Map of the dataset folders"></p>
<p align="center">
</p>
| Folder | Contents |
| --- | --- |
| [`transcripts/`](transcripts) | The complete session histories for the three models (JSON), with plain-text renderings of the archived chat-transcript PDFs. |
| [`data/`](data) | The master dataset and its derived versions (extracted features, classified and completed sets, and the APD, cluster, embedding, rater-agreement, and typology extensions), with the session-level and turn-level label files. |
<p align="center"><img src="assets/divider.png" alt="" width="80%"></p>
## 📚 Citation
Please cite the paper. Citation metadata is provided in [CITATION.cff](CITATION.cff).
```bibtex
@misc{hamidavi2026gaslighting,
author = {Hamidavi, Abbas},
title = {Algorithmic Gaslighting: A Taxonomy and 15-Session Case
Series of Reality-Distortion Behaviour in Frontier LLMs},
year = {2026},
note = {Position paper submitted to the CALE-AI Workshop at
AAAI-27 (Submission 4). Longer version: Research Square
rs-10386085}
}
```
<p align="center"><img src="assets/divider.png" alt="" width="80%"></p>
## 🔗 Access to the article
- **Longer version (Research Square, open access):** [researchsquare.com/article/rs-10386085](https://www.researchsquare.com/article/rs-10386085)
- **OSF deposit:** [osf.io/qsvty](https://osf.io/qsvty)
- **Venue:** position paper submitted to the CALE-AI Workshop at AAAI-27 (Submission 4).
<p align="center"><img src="assets/divider.png" alt="" width="80%"></p>
## ⚖️ Licence
The contents of this repository are released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0, see [LICENSE](LICENSE)).
The transcripts are outputs of publicly available AI systems, generated for this study in response to scripted fictional scenarios. They contain no personal data.
<p align="center">
</p>
<p align="center"><img src="assets/footer.png" alt="Abbas Hamidavi, Machine Psychology research programme"></p>