This research introduces a comprehensive methodology for consistent psychometric testing of large language models, addressing critical reproducibility concerns in AI evaluation.

• Identifies fundamental instabilities in how LLMs respond to psychological assessments
• Develops a unified testing framework to standardize evaluations across different models
• Establishes protocols for reliable psychometric measurement in AI systems
• Proposes solutions to reduce output variability and improve test validity

From a security perspective, standardized psychological profiling of AI systems is essential for alignment verification and identifying potentially problematic behaviors before deployment.

R.U.Psycho? Robust Unified Psychometric Testing of Language Models