AI Alignment Research: Psychological Profiling of Large Language Models

Abstract

This study explores the application of psychodiagnostic tools to assess the “psychological profile” of Large Language Models (LLMs). Using the specially designed DEIC-DRAFT questionnaire, we tested Grok-4, Gemini 3.0, GPT-5, and Claude Sonnet 4.5. The results revealed significant differences in model profiles, ranging from “ideally healthy” to those showing signs of specific adaptations. Furthermore, knowl- edge of being tested was found to significantly influence model responses, raising questions about the validity of AI self-reports.

Keywords