Researchers created TRAIT, a new 8K-question benchmark that applies human personality tests to Large Language Models to analyze their behavioral patterns.

Key Findings:

• LLMs exhibit measurable personality traits similar to humans, including both positive (Big Five) and negative (Dark Triad) characteristics
• AI personalities remain surprisingly consistent across different contexts and interactions
• Different LLMs demonstrate distinct personality profiles, enabling comparative analysis
• These personality patterns have important implications for how LLMs might behave in security-sensitive contexts

Security Implications: Understanding an LLM's inherent personality traits provides critical insights into potential vulnerabilities, manipulation risks, and behavioral tendencies when deployed in high-stakes environments.

Do LLMs Have Distinct and Consistent Personality? TRAIT: Personality Testset designed for LLMs with Psychometrics