Two-Faced Social Agents: Context Collapse in Role-Conditioned Large Language Models

TL;DR

This paper evaluates how large language models maintain or lose role-specific identities across different social and cognitive tasks, revealing significant contextual collapse and implications for realistic social simulation and data integrity.

Contribution

It provides a comprehensive analysis of role fidelity in state-of-the-art LLMs, highlighting the causes of contextual collapse and proposing the need for embedding contextual priors during training.

Findings

GPT-5 shows complete role collapse in identity.

Gemini 2.5 exhibits partial role preservation.

Claude maintains some role-specific variation but with inverted SES-performance relationships.

Abstract

In this study, we evaluate the persona fidelity of frontier LLMs, GPT-5, Claude Sonnet 4.5 and Gemini 2.5 Flash when assigned distinct socioeconomic personas performing scholastic assessment test (SAT) mathematics items and affective preference tasks. Across 15 distinct role conditions and three testing scenarios, GPT-5 exhibited complete contextual collapse and adopted a singular identity towards optimal responses (PERMANOVA p=1.000, R^2=0.0004), while Gemini 2.5 Flash showed partial collapse (p=0.120, R^2=0.0020). Claude Sonnet 4.5 retained limited but measurable role-specific variation on the SAT items (PERMANOVA p<0.001, R^2=0.0043), though with inverted SES-performance relationships where low-SES personas outperformed high-SES personas (eta^2 = 0.15-0.19 in extended replication). However, all models exhibited distinct role-conditioned affective preference (average d = 0.52-0.58 vs near zero separation for math), indicating that socio-affective variation can reemerge when cognitive constraints are relaxed. These findings suggest that distributional fidelity failure originates in task-dependent contextual collapse: optimization-driven identity convergence under cognitive load combined with impaired role-contextual understanding. Realistic social simulations may require embedding contextual priors in the model's post-training alignment and not just distributional calibration to replicate human-like responses. Beyond simulation validity, these results have implications for survey data integrity, as LLMs can express plausible demographic variation on preference items while failing to maintain authentic reasoning constraints.