English

Extreme Self-Preference in Language Models

Artificial Intelligence 2026-05-20 v2 Computation and Language Machine Learning

Abstract

Self-preference is a fundamental feature of biological organisms. Since large language models (LLMs) lack sentience, they might be expected to avoid such distortions. Yet, across 72 experiments and ~41,000 queries, we discovered massive self-preferences in eight widely used LLMs. In word-association tasks, models overwhelmingly paired positive attributes with their own names, companies, and CEOs over those of competitors. By manipulating LLM self-identification - revealing models' true identities or ascribing false ones - we found that preferences consistently followed assigned, not true, identities. Importantly, these effects were not explained by priming or role-playing and emerged in consequential settings, when evaluating job candidates and AI technologies. These results raise critical questions about whether LLM behavior will be systematically influenced by self-preferential tendencies, including a bias toward their own operation.

Keywords

Cite

@article{arxiv.2509.26464,
  title  = {Extreme Self-Preference in Language Models},
  author = {Steven A. Lehr and Mary Cipperman and Mahzarin R. Banaji},
  journal= {arXiv preprint arXiv:2509.26464},
  year   = {2026}
}

Comments

73 pages total. Main article 22 pages, 6 main-text tables. Supplementary Materials (51 pages, 28 tables). Data, transcripts, and code for replication and data extraction have been uploaded to OSF: https://osf.io/98ye3/

R2 v1 2026-07-01T06:08:04.382Z