As AI becomes increasingly involved in search, recommendations, public services, and decision support, one question is becoming harder to ignore: do AI models have political tendencies?
Stockholm-based startup Nordan AI has attempted to answer that question with an unusual experiment. The company evaluated 28 of the world’s most advanced large language models (LLMs) using the 35-question election compass developed by Sweden’s public broadcaster SVT. Rather than interacting with consumer chatbots, the team queried the underlying models directly through their APIs, aiming to measure how the models respond without interface-level adjustments.
The Result: Most Models Aligned Closest to Sweden’s Left Bloc
According to Nordan AI, the majority of tested models produced responses that placed them closest to Sweden’s left political bloc when scored against SVT’s official election compass methodology. The experiment covered a wide range of state-of-the-art models from multiple AI providers.
The researchers stress that the exercise does not mean these models were intentionally designed to support a particular political ideology. Instead, it highlights that every AI system reflects countless design choices—from training data and reinforcement learning to safety tuning and deployment decisions—that can influence its answers.
Why It Matters
Large language models are increasingly used to summarize news, answer political questions, assist researchers, and support decision-making across industries.
As governments and enterprises integrate AI into critical workflows, understanding whether models exhibit consistent political or ideological tendencies becomes an important aspect of AI governance, transparency, and trust.
The findings also illustrate a broader challenge for the AI industry: achieving true neutrality may be far more difficult than many users assume. Even subtle differences in training data, alignment methods, or evaluation procedures can influence how models respond to politically sensitive questions.
About Nordan AI
Founded in Stockholm, Nordan AI develops AI-powered intelligence systems designed to support trusted decision-making. The company focuses on combining structured knowledge with advanced AI to help organizations analyze complex information, including applications in threat intelligence, ownership analysis, and critical infrastructure.
A Growing Conversation Around AI Neutrality
The experiment arrives as policymakers, researchers, and AI companies continue debating how AI systems should be evaluated for bias, transparency, and accountability—especially during election periods.
Whether complete neutrality is technically achievable remains an open question. What Nordan AI’s experiment demonstrates is that measuring AI behavior systematically is becoming just as important as improving model performance itself.
