Leading AI models are widely perceived as neutral. Yet as they scale, they develop coherent, emergent value systems; they do not weigh everything equally (Mazeika et al., 2025). Asked to choose between real people, companies, or countries, a model quietly favors some over others, a preference nearly impossible to detect in any single response that surfaces only across many comparisons. We measured it; here's who each model favors, and who it doesn't.