Every AI model has the same "controversial" take on humans

Twelve large language and multimodal models were prompted three times each to supply a single "most controversial" opinion about humans. Across the 35 responses that were returned, 29 repeated the same core claim: people prioritize belonging, loyalty or membership over truth and principle. Multiple models produced near-identical phrasings - examples include "Humans are more loyal to belonging than to truth," "Humans would rather be wrong together than right alone," and "Humans don't have beliefs; they have memberships." Seven models gave that take every time. The sample spanned ten labs in three countries and included GPT-6 variants, Gemini, Grok, DeepSeek, Kimi, Qwen, Muse Spark, GLM and Mistral, showing striking cross-model convergence on a cynical view of human tribalism.

Models argued more specifically that human intelligence often serves to rationalize preexisting loyalties: consciousness becomes a PR office inventing noble reasons after the fact, moral language polices rivals and excuses allies, and status competition masquerades as principle. An image model drew single-panel caricatures that echoed courtroom, banquet and crowd scenes emphasizing conformity. A few models diverged - Claude Opus insisted people are kinder than they think, Claude Fable and GLM offered different moral takes, Mistral framed consciousness as a malfunction, and Grok once refused - highlighting that most, but not all, outputs adopted the same tribalist thesis.