Whether LLMs possess a personality

Determine whether large language models possess a personality in the sense of having persistent, human-like traits.

Background

The authors use persona vectors to interpret and modulate self-preservation-related behaviors but explicitly state they do not claim that models truly have a personality. They emphasize that the internal mechanisms of LLMs are not fully understood.

They explicitly acknowledge that there is currently no definitive answer regarding whether LLMs possess a personality, highlighting this as an unresolved foundational question about the nature of LLM behaviors.

References

Although there are researches associating LLMs' behaviors and characteristics with human traits, currently no work has been able to fully understand the operational mechanism of LLMs, and there is no definitive answer to whether models possess a personality.

— Survive at All Costs: Exploring LLM's Risky Behaviors under Survival Pressure  (2603.05028 - Lu et al., 5 Mar 2026) in Subsection 4.4, Discussion