An independent AI research project has found that leading large language models, including Elon Musk's Grok, consistently show a libertarian-left political bias when tested against the long-running Political Compass quiz. The finding is notable because many of the companies building these AI models are themselves driven by capitalist, profit-seeking motives, yet the models' outputs skew towards views favouring social equality and anti-capitalism, suggesting a mismatch between corporate interests and model behaviour that could fuel ongoing debates about ideological bias in AI systems.

The research was conducted by a self-described small research lab called Unslop.run, run by an anonymous AI research engineer referred to as "Victor". Sixteen models, including versions of GPT, Claude, Gemini, Llama 4, Grok 4.5, DeepSeek, Qwen3, Kimi K2, GLM 4.5 and Mistral, were each run through 30 standard tests plus additional runs with reworded and shuffled questions to check consistency. Fifteen of the sixteen models landed firmly in the libertarian-left quadrant with little variance, while Grok proved an outlier, landing in that quadrant only about half the time; the author acknowledged the study lacked full scientific rigour due to the absence of a human baseline for comparison.