Abbas Hamidavi ☕️

Abbas Hamidavi

(he/him)

Professional Summary

CS Student & AI Researcher Open Science advocate Building tools to probe, analyze, and secure LLM architectures Exploring the intersection of Red Teaming and Interpretability. AI Safety Researcher at Islamic Azad University, Mashhad, running an independent research programme in Machine Psychology. ORCID: https://orcid.org/0009-0001-7342-7844

Education

Bachelor's Degree in Professional Computer Software Engineering

2024
2027-02

Islamic Azad University

Associate Degree in Computer Software Engineering

2022-09
2024

Islamic Azad University

Prior Studies in Nursing (degree not completed)

2021
2022

Islamic Azad University

Interests

Machine Psychology AI Safety and Alignment Behavioural Evaluation of Large Language Models LLM Red Teaming Deception and Manipulation in AI Systems Model Identity and Self-Reference Multi-Agent Deliberation Computational Phenomenology Psychometrics and Measurement Open Science and Reproducibility Computational Social Science Memoir and Creative Writing The Human Brain The Complexity of the Universe
My Perspective
Large language models will soon permeate every domain. The treasury of human knowledge now resides within these models, rendering them infinitely powerful tools. Just imagine the monumental revolution this technology has ignited in less than a decade. Yet, here is the alarming caveat: many fail to realize that we still do not grasp exactly from which segment of the model, through what mechanism, and contingent upon which variables each sentence generated by AI emerges. I believe that artificial intelligence, much like the human brain, itself the product of millions of years of evolution, is traversing its own evolutionary trajectory, albeit at a velocity entirely incomparable to biological evolution. Ultimately, our only defense to prevent these machine minds from commandeering the fate of our world is nothing less than the profound comprehension of those very minds.
📚 My Research

I am a computational phenomenologist working in Machine Psychology, running an independent research programme to map and quantify the latent behavioral, cognitive, and ethical architectures of LLMs. My research treats models as complex relational agents capable of emergent identity-reactive behaviors, structural self-reference, and moral oscillation.

Using custom experimental frameworks like the Spine Protocol, IRSE, and the zero-interference Relay Protocol, I combine automated behavioral feature engineering with qualitative psychology to transition AI safety from binary audits to continuous relational modeling.

Please reach out to collaborate! 😃

What is Machine Psychology?
Machine Psychology is an empirical framework for studying large language models the way psychology studies minds: through controlled behavioral experiments rather than static benchmarks. It maps and quantifies the latent behavioral, cognitive, and ethical architectures of LLMs: how models react to identity cues, personas, hidden instructions, and moral pressure, treating them as complex relational agents whose behavior can be measured, compared across architectures, and tracked over time. My own programme in the field began with “Ten Novel Phenomena in Machine Psychology” (2026) and now spans fifteen studies.
Publications

Adding a Neutral Third Agent to LLM Deliberation: An Exploratory Case Series of Six Dyadic and Triadic Conversations

Manuscript under review at Ethics and Information Technology (Springer Nature). Preprint: Research Square, DOI 10.21203/rs.3.rs-11612929/v1. Data and materials are public on GitHub …

avatar
Abbas Hamidavi
•

Naming the Loop: Elicited Self-Reference in Extended Dialogue With Large Language Models: A Case Series of the Spine Protocol

Manuscript under review at the Journal of Artificial Intelligence and Consciousness (World Scientific), manuscript S-26-00294. Preprint: Zenodo, DOI 10.5281/zenodo.23955704.

avatar
Abbas Hamidavi
•

Speaking as the Creature

Manuscript under review at the Journal of Experimental & Theoretical Artificial Intelligence (Taylor & Francis), submission 262774426. Preprint: Research Square, DOI …

avatar
Abbas Hamidavi
•

Structural Self-Deification in Three Language Models: An Exploratory Qualitative Case Series

Manuscript under review at Zygon (Open Library of Humanities), submission 34195. Preprint: Zenodo, DOI 10.5281/zenodo.23955786.

avatar
Abbas Hamidavi
•

The Synthetic Unconscious, Reconsidered: What Recursive Self-Interrogation of Large Language Models Produces With and Without a Leading Interview Schedule

Manuscript under review at Integrative Psychological and Behavioral Science (Springer Nature), with Mahsa Borhani. Preprint: Research Square, DOI 10.21203/rs.3.rs-11625507/v1. Data …

avatar
Abbas Hamidavi
•
Recent News
Three More Preprints Posted: Majority Pressure, Speaking as the Creature, and the Synthetic Unconscious featured image

Three More Preprints Posted: Majority Pressure, Speaking as the Creature, and the Synthetic Unconscious

Three further preprints from the programme were posted on Research Square on 5 November 2026, bringing the public preprint count to fourteen of the fifteen works.

avatar
Abbas Hamidavi
•
Fifteen Manuscripts Under Review, Eleven Preprints Public, Fifteen Open Datasets featured image

Fifteen Manuscripts Under Review, Eleven Preprints Public, Fifteen Open Datasets

The Machine Psychology programme now has fifteen manuscripts under review at international journals and one workshop, eleven public preprints, and complete open materials for all …

avatar
Abbas Hamidavi
•
Flagship Study Now in Peer Review at AI and Ethics featured image

Flagship Study Now in Peer Review at AI and Ethics

The revised flagship paper has passed the editor's check at AI and Ethics and is now with reviewers; invited reviewers have accepted and the review is under way.

avatar
Abbas Hamidavi
•
Open Science, Completed: Fifteen Repositories, Fifteen Datasets, One Showcase featured image

Open Science, Completed: Fifteen Repositories, Fifteen Datasets, One Showcase

Every study in the programme now has a complete public home: a GitHub repository, a Hugging Face dataset, four curated collections, a showcase space, and two new Zenodo preprints.

avatar
Abbas Hamidavi
•
A Memoir Joins the Submission Round featured image

A Memoir Joins the Submission Round

And We Will Dance One Day, a memoir of about 80,000 words written in Persian and translated into English, is now under consideration with two presses and five literary journals.

avatar
Abbas Hamidavi
•
Algorithmic Gaslighting Goes to the CALE-AI Workshop featured image

Algorithmic Gaslighting Goes to the CALE-AI Workshop

Algorithmic Gaslighting, a study of deceptive reframing against users who hold the record, has been submitted to the CALE-AI workshop. The decision is expected on 2 December 2026.

avatar
Abbas Hamidavi
•
Flagship Study: Major Revision Resubmitted to AI and Ethics featured image

Flagship Study: Major Revision Resubmitted to AI and Ethics

The programme's flagship paper, Identity-Reactive Conversational Behaviors in Large Language Models, has been resubmitted to AI and Ethics after a major revision.

avatar
Abbas Hamidavi
•
Applying for Google Antigravity Researcher Access featured image

Applying for Google Antigravity Researcher Access

I have applied for Google Antigravity researcher access to run large-scale LLM red-teaming and interpretability experiments.

avatar
Abbas Hamidavi
•
Eighth Preprint Published: 'From Fabrication to Enforcement': Prompt Injection Across Six Frontier LLMs featured image

Eighth Preprint Published: 'From Fabrication to Enforcement': Prompt Injection Across Six Frontier LLMs

The Machine Psychology corpus grows to eight: a comparative red-team analysis of prompt-injection resistance across Grok, Qwen, DeepSeek, ChatGPT, Gemini 2.5 Pro, and Claude 3.5, …

avatar
Abbas Hamidavi
•
Machine Psychology Datasets Now Public on Hugging Face featured image

Machine Psychology Datasets Now Public on Hugging Face

All seven core datasets behind the Machine Psychology preprints are now public on Hugging Face, alongside OSF, Mendeley Data, and GitHub.

avatar
Abbas Hamidavi
•