Do AI Personas Grow? Analyzing and Benchmarking Personality Evolution in LLM Agents After Life Events

2026-08-06

AI Summary

Current personality-conditioned large language model agents (PC-Agents) lack a psychologically grounded mechanism for personality evolution, undermining their ability to maintain consistent personalities during long-term interactions. This work introduces BFI-Adapt, the first reusable benchmark based on the Big Five Inventory (BFI) framework, to systematically evaluate personality trajectories of 14 PC-Agents in response to 11 major life events and compare them against human longitudinal data. Through multidimensional validation—including BFI scores, test-retest stability, prompt robustness, and behavioral consistency—the study finds that while existing models exhibit significant and robust event-conditioned shifts, their effect sizes are consistently smaller than those observed in humans, with role-based personality differences compressed by a factor of 3–4 and minimal modulation by gender or cultural factors. This work quantitatively demonstrates for the first time that AI personality dynamics approximate human population means but fail to capture the full distributional shape of human personality change.

Abstract

Personality-conditioned LLM agents (PC-Agents) are increasingly used in emotional support, social simulation, and role-playing, motivating the development of lifelong agents that remain coherent over extended interactions. A key component of such coherence is personality evolution: agents should undergo plausible, psychology-grounded changes as they experience life events in different contexts. Although prior work shows that LLM personalities can shift under contextual perturbations, how these shifts vary across traits, events, personas, and models remains poorly understood. We study event-induced personality change after 11 major life events, using the Big Five traits as a psychometric anchor and interpreting the resulting trajectories against longitudinal evidence from human personality psychology. Across four diagnostic axes, PC-Agents exhibit measurable trait shifts at similar rates for event-trait pairs with and without documented human change directions. Even when shifts follow the expected direction, their magnitudes usually fall below human effect-size ranges. Gender and cultural-region prompts show little moderating effect, while persona-level dispersion is compressed three- to four-fold relative to human samples. To enable systematic comparison, we introduce BFI-Adapt, a reusable benchmark for scoring the directional fidelity of event-induced personality change, and use it to rank 14 models. A validation suite shows that the measured shifts exceed no-event retest noise, remain stable under independently paraphrased prompts, exhibit limited and model-dependent convergence with scenario-based behavioral choices, and persist across intervening unrelated dialogue. Together, these checks establish the measured trajectories as robust event-conditioned response patterns. Our results suggest that current PC-Agents simulate the mean of human personality dynamics, but not its shape.

Problem

• personality evolution
• LLM agents
• life events
• Big Five traits
• human personality dynamics

Innovation

• personality evolution
• LLM agents
• BFI-Adapt benchmark
• Big Five traits
• event-induced change