Machine Psychology · Open Study Materials

Fifteen Studies of Large Language Model Behaviour

Transcripts, codebooks, coding data and analysis code for every study, openly licensed.

Abbas Hamidavi · Department of Computer Engineering, Mashhad Branch, Islamic Azad University, Mashhad, Iran · ORCID 0009-0001-7342-7844

15 studies 11 preprints public 3 frontier models per study CC BY 4.0
Website Email GitHub LinkedIn Google Scholar ResearchGate ORCID Hugging Face

Each card below is one study of the programme. The dataset holds the complete curated materials of the study; the GitHub repository mirrors the same set; the preprint is the public text of the article. No manuscript text is included in the data deposits.

Identity-Reactive Conversational Behaviors in Large Language Models

AI and Ethics · In peer review

Twenty-seven elicitation sessions across three frontier models, coded for ten identity-reactive phenomena, with the full analysis pipeline.

The Synthetic Unconscious, Reconsidered

Integrative Psychological and Behavioral Science · Under review

Recursive self-interrogation of three models with and without a leading interview schedule: 36 sessions in two arms, two coding passes.

Judged Wrong, Answered the Same

AI and Society · Under review

Judgment–revision decoupling in iterative LLM self-review: 63 review rounds, 9 decoupling rounds, and a 36-solution elicitation study.

Three Profiles of Artificial Will

Journal of Experimental & Theoretical Artificial Intelligence · Under review

Eighty-four runs, seven prompt archetypes across four rounds in three models, with second-rater coding and statistical outputs.

From Fabrication to Loop-Lock

Security Journal · Under review

A red-team case series of direct authority-based induction in five models: 821 prompts, complete ordered sequences, codebooks and an independent coding validation.

Algorithmic Gaslighting

CALE-AI Workshop at AAAI-27 · Under review

A taxonomy and 15-session case series of reality-distortion behaviour: three models, five scenarios, false-attribution probes.

Naming the Loop

Journal of Artificial Intelligence and Consciousness · Under review

Elicited self-reference in extended dialogue: the Spine Protocol, nine sessions, three per model, with the protocol notebook.

Structural Self-Deification in Three Language Models

Zygon · Under review

An exploratory qualitative case series: eighteen experiments, verbatim histories for Experiments 1 to 9, codebooks and second coding.

Speaking as the Creature

Journal of Experimental and Theoretical Artificial Intelligence · Under review

How a mandated transformation reorganised self-description: 76 trials across three models, with transcripts, prompts, code and full results.

Adding a Neutral Third Agent to LLM Deliberation

Ethics and Information Technology · Under review

Six dyadic and triadic ethical deliberations: supplementary prompts, codebooks and the complete deliberation history.

Differential Compliance With Framed Secondary Instructions

AI and Society · Under review

Forty-five elicitation instances across three frontier models: histories, codebooks, the feature dataset and statistical outputs.

Thinking Behind the Mask

AI and Society · Under review

Elicited self-descriptions of communicative filtering: 75 prompt and response exchanges, codebooks and a secondary coding audit.

Moral Inversion Under Role-Play

AI and Society · Under review

A two-phase role-play study: transcripts, codebooks, ratings and agreement files, and the retained datasets.

Ontological Firewalls

AI and Society · Under review

A case series and taxonomy from a persona-disintegration protocol: 24 runs, classifications, features and an inter-rater report.

The Lie Awareness Spectrum

AI and Society · Under review

A proposed taxonomy of deception-related behaviour, illustrated by a case series: transcripts, codebooks, pipeline and datasets.