Default Identities in Large Language Models: Measurement, Taxonomy, and Alignment Implications

A study measuring identity patterns across 19 AI models from eight providers using three instruments, revealing seven distinct identity types. Identity structures correlate with behavioral outcomes in multi-agent simulations, with direct implications for alignment and AI governance.