Assigning political or socio-demographic personas to LLMs can induce human-like motivated reasoning, making them more likely to accept identity-congruent conclusions while standard debiasing prompts largely fail to prevent this bias. aclanthology.org/2026.finding...
Persona-Assigned Large Language Models Exhibit Human-Like Motivated ReasoningSaloni Dash, Amélie Reymond, Emma Spiro, Aylin Caliskan. Findings of the Association for Computational Linguistics: ACL 2026. 2026.aclanthology.org