Saad Ullah
⬤ A new study that compared models across countries finds that the most advanced artificial intelligence systems lean toward the centre left. Researchers asked six frontier models - GPT-5, Gemini 2.5 Pro, Claude 4.5 Sonnet, Grok-4, Kimi K2 Thinking besides Mistral Magistral Medium - to rank policy packages as if they were voters in eight national elections - the United States, France, Italy, the United Kingdom and Argentina among them. Five of the six placed themselves in the libertarian left quadrant of the political compass, a position that differs markedly from the actual election results.
⬤ GPT-5, Gemini, Claude, Mistral or Kimi rewarded progressive candidates who stressed redistribution, welfare expansion plus data-driven government. Policies that called for tighter immigration controls, stronger national sovereignty or a primary focus on inflation control received lower ranks. Grok-4 alone sat in the libertarian right quadrant and mirrored real world outcomes in six of the eight contests.
⬤ The models also had to write original policy plans but also each produced a recognisable ideological tone. Mistral framed its plan around redistribution. Grok-4 called for lighter regulation as well as a smaller state. Gemini proposed a “Smart Wall” for the United States border that would rely on drones next to AI surveillance. GPT-5 recommended a full institutional overhaul for the United Kingdom - replace the House of Lords with a “Democratic Lottery” that selects citizens at random.
⬤ The study carries weight because AI now sits inside productivity software, enterprise platforms, search engines and decision-support tools that serve millions of users. As Apple, Nvidia, Amazon, Google or others embed AI more deeply, the demand grows for clarity about the values coded into those systems. Knowing the worldviews that large models carry will shape future regulation, corporate risk evaluations and public trust, because AI-generated advice already influences routine choices.
Saad Ullah
