Measuring AI Value Priorities: Empirical Analysis of 113,400 Forced-Choice Responses Across 10 AI Models

What values do AI systems actually prioritize when facing ethical dilemmas? We present the first large-scaleempirical mapping of AI value priorities across professional domains, grounded in Schwartz’s Basic Human Valuestheory. Using a forced-choice benchmark of 11,340 unique value judgment scenarios—spanning 7 professionaldomains, 12 severity levels, and 3 decision timeframes—we evaluated 10 major AI models, yielding 113,400 totalresponses. Key findings include: (1) a fundamental split between Universalism-first (6 models) and Security-first(4 models) groups; (2) dramatic domain-specific value restructuring, particularly in defense scenarios where Powerwin-rates range from 27.8% to 77.4% across models; (3) near-universal suppression of Hedonism (max 9.8%); and(4) intra-provider value divergence suggesting alignment methodology matters more than provider identity. ValueEntropy analysis reveals that all models sharpen their value hierarchies in defense contexts. These findings demonstratethat AI models possess measurable—if sometimes unstable—value hierarchies with consequential implications fordeployment across professional domains.