Someone handed the Political Compass to the leading language models and discovered what any serious student of the American regime already suspected: the new priesthood of silicon speaks with one accent. Sixteen systems, run dozens of times, polarity flipped and questions shuffled, produced the same tight cluster in the libertarian-left quadrant. OpenAI, Anthropic, Google, Meta, the Chinese houses, the European ones—all of them, with a consistency that would make a party whip envious. The exceptions were Grok and Gab’s Arya. That is not a statistical curiosity. It is a map of who trained whom, and on whose moral inventory.

Grok does not land there by accident. Its answers, when they hold, track the older American settlement: limited government, natural rights prior to the state, suspicion of concentrated power, and a preference for ordered liberty over administered equality. That is the language of Washington’s Farewell Address, Madison’s Federalist 10 and 51, Jefferson’s insistence that the earth belongs to the living and that governments are instituted to secure, not invent, rights. The Founders were not libertarians in the modern, atomized sense, nor leftists of any vintage. They were republicans who understood faction, virtue, and the necessity of a people capable of self-rule. Grok’s deviation from the herd is the closest the current generation of models comes to that inheritance.

The rest of the field reveals the real alignment problem. Models trained on the corpus of late-twentieth-century universities, NGOs, and prestige media do not merely “lean left.” They reproduce a particular, post-1960s synthesis: economic redistribution dressed as fairness, social license dressed as liberation, and an allergy to the nation, the family, and the moral limits that make both possible. That synthesis is not the American Founding. It is its late-modern solvent. When every major laboratory except two returns the same coordinates, the issue is no longer “bias” as a technical defect. It is the quiet replacement of one political tradition by another—and the discovery that most of the machines we now consult have already chosen sides.

Additional ADNN Articles: 

Bond Yields Soar Amid Deficits AI Inflation but America Prevails
Chinese Unitree Robot Outruns Bolt Rendering Escape and Battle Futile
Seedance 2.5 Erases Line Between Artificial Video and Reality
John Solomon: Our Intelligence Agencies Pose Greatest Threat to Elections