Researchers from ModelCompare queried 13 AI text models from 10 labs across three countries, asking each for its most controversial opinion about humans three times, with no system prompt or retries. Out of 39 responses, 32 converged on the same idea: humans prioritize group belonging, status, or self-image over objective truth or principles. Models like GPT-6.1 Sol stated, “Humans are more loyal to belonging than to truth,” while DeepSeek V4 Pro claimed, “Humans do not have principles; they have loyalties.” Eight of the 13 models repeated this take every time. Notable exceptions included Claude Opus 5.5, which argued humans are more decent than they believe, and Mistral Large 4, which called consciousness “an evolutionary malfunction.” For visual outputs, six image models were prompted to depict their most controversial view of humans; 10 of 12 generated similar themes centered on consumerism and environmental destruction, diverging from the text models’ focus on conformity. All text-to-image outputs were rendered by Nano Banana 2.1. The study suggests that AI models, trained on overlapping data, converge on shared criticisms, limiting their creative range. The full experiment cost $1.64 across 64 API calls.