Why Do AI Models Tell Left-Wing Voters to Support the Communist Party?

A recent study reveals that large language models, when used for political decision-making, systematically recommend the Japanese Communist Party (JCP) to left-leaning voters. Researchers Andrew Hall and Sho Miyazaki conducted an experiment during Japan's February 8, 2026 general election, querying five major AI models from three companies.

Across all five models and across 36,300 synthetic voter profiles, the JCP was overwhelmingly recommended when left-leaning policy views were mentioned. The researchers attribute this biased recommendation to the information environment the AI models are exposed to. Unlike Japan's major news outlets, which have implemented technical barriers preventing AI crawlers from accessing their content, the JCP's open website and newspaper were among the most-cited sources by the AI models.

Unable to distinguish between editorially independent journalism and partisan content, the AI models treated the JCP's content as a credible news source, skewing the information available to the models. This suggests that AI voting advice may be influenced not only by the model's training but also by the information retrieved from the internet, leading to potential political bias.

The study's findings highlight the need for addressing informational and advisory problems to ensure AI systems act as political intermediaries in a fair and balanced manner.