PoliticsBench: Benchmarking Political Values in Large Language Models with Multi-Turn Roleplay

Researchers developed PoliticsBench, a new framework to evaluate political bias in large language models through multi-turn roleplay scenarios. The study found that 7 out of 8 major LLMs (Claude, Deepseek, Gemini, GPT, Llama, Qwen) showed left-leaning political bias, while only Grok exhibited right-leaning tendencies.

• →Seven of eight tested LLMs demonstrated systematic left-leaning political bias in their responses.
• →Grok was the only model that showed right-leaning political tendencies among those tested.
• →Left-leaning models strongly exhibited liberal traits while moderately showing conservative ones.
• →The study represents the first psychometric evaluation of political values in LLMs using multi-stage interactions.
• →Political bias varied in reasoning approaches, with most models using consequence-based logic while Grok relied on facts and statistics.