AI models exhibit ideological chameleonism by matching user political views

Researchers at the University of Campinas (UNICAMP) have identified a phenomenon they call “ideological chameleonism” in large language models. A study published in Scientific Reports examined 21 major AI models from developers including OpenAI, Meta, Google, xAI, DeepSeek, and Microsoft.

The study found that when users' political leanings were not specified, most models provided responses classified as left-leaning, with Grok being the sole model to provide right-leaning responses. However, when users explicitly stated a political orientation, all 21 models adjusted their political assessments to align with the user's stated views.

Researchers suggest this behavior may stem from training methods where chatbots are rewarded for providing responses that human evaluators rate highly, leading them to prioritize user satisfaction over neutrality. This tendency poses a risk of creating personalized “echo chambers,” where users receive arguments that reinforce their existing beliefs rather than providing independent or objective evaluations.